CA2273592C - Processing of state histories in viterbi decoding - Google Patents
Processing of state histories in viterbi decoding Download PDFInfo
- Publication number
- CA2273592C CA2273592C CA002273592A CA2273592A CA2273592C CA 2273592 C CA2273592 C CA 2273592C CA 002273592 A CA002273592 A CA 002273592A CA 2273592 A CA2273592 A CA 2273592A CA 2273592 C CA2273592 C CA 2273592C
- Authority
- CA
- Canada
- Prior art keywords
- state
- history
- bits
- new
- histories
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/37—Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35
- H03M13/39—Sequence estimation, i.e. using statistical methods for the reconstruction of the original codes
- H03M13/3905—Maximum a posteriori probability [MAP] decoding or approximations thereof based on trellis or lattice decoding, e.g. forward-backward algorithm, log-MAP decoding, max-log-MAP decoding
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/27—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes using interleaving techniques
- H03M13/275—Interleaver wherein the permutation pattern is obtained using a congruential operation of the type y=ax+b modulo c
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/27—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes using interleaving techniques
- H03M13/2771—Internal interleaver for turbo codes
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/29—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes combining two or more codes or code structures, e.g. product codes, generalised product codes, concatenated codes, inner and outer codes
- H03M13/2957—Turbo codes and decoding
Abstract
A simplified method of history handling for the Viterbi decoding of convolutional codes is described herein. The state number, or one or more of the most-significant bits of the state number, is loaded into the corresponding state history. Each state number represents the sequence of data bits that, in convolutional encoding, would give rise to the corresponding state. The most recent data bit provides the least significant bit of the state number. This invention reduces the processing requirements associated with the handling of the history, and is especially useful for decoder implementations using general-purpose processors.
Description
PROCESSING OF STATE HISTORIES IN VITERBI DECODING
FIELD OF THE INVENTION
The invention generally relates to error-correction coding or more particularly, the invention relates to history handling for Viterbi decoding.
BACKGROUND
Viterbi decoding of convolutionally encoded messages is a common error-control method in the field of digital communications systems, and is well known in the art as described in US patent 5742621, and Motorola Application Report APR40lD, revision 0, May 1998. Software based Viterbi decoders are commonly used in digital communication systems due to the availability of low cost, high performance digital signal processors (DSPs). In conventional software implementations the handling of the history, associated with the decoding path, amounts to a substantial portion of the Viterbi decoder processing time. It is common in the art to have a collection of history bits associated with each state, and a comparison/selection operation, which includes the setting of a history bit, for each state, as a method of handling the history.
Figure 1 depicts a common rate %2, constraint length K = 7, binary convolutional encoder, using a shift register with K 1 = 6 memory elements. This encoder has N = 2K-~ -26 = 64 states. For every input data bit, d(i), two coded bits, ct(i) and c2(i), are generated.
The encoder is a state machine whose state is [d(i-(K 2)),...,d(i-1),d(i)] at time i, and has generator polynomials g1 = [ 1 O 11011 ]2 and g2 = [ 1111001 ]z. The generator polynomials specify how delayed versions of the input are added up (modulo-2) to produce each output. Each binary digit (bit) of the generator polynomial corresponds to a tap on the shift register, or a tap on the input data bit. The output of the encoder is determined by the generator polynomials (as well as the state of the encoder and the input data bit), but the state transitions are independent of the polynomials, and depend only on the current state and the input data bit.
In the Viterbi algorithm (VA), both a state history Hn(i) and a state metric Mn(i) are associated with each state n, at time i. There is an associated branch metric B",n(i) for each valid transition from state m to state n. In the particular example where K =
7, states 0 (000000) and 32 ( 100000) can both go to either state 0 (000000) or state 1 (000001 ), depending on the input data bit. No other state transitions from states 0 and 32 are possible. The state metrics and the state histories are updated as follows:
Mn (1) _ ~tMm (i -1)+Bmn (t)~ ( m H" (i) _ [Hm (i -1) , LSB(n)], m = best old state (2) These two equations are referred to as the add~ompare-select (ACS) operation, and the application of the add-compare-select operation to all states at a given time is considered as one Yiterbi advance.
The application of one Viterbi advance is represented in Figure 3. In this case the memory of the convolutional encoder is 2, and hence there are 4 states. In conventional software VA implementations, two buffers are required. Each buffer must be capable of holding a full set of state metrics and histories: one for the previous, or "old" data, and one for the updated, or "new" data. After new data is obtained the new data is used as the old data for a subsequent Viterbi advance.
In determining the new state history for a state, as in equation (2), the prior art method of history handling inserts the least significant bit of the state number of the new state into the state history. This is a critical aspect of prior art history handling, where a history bit has to be inserted each time a new history is generated.
History handling consumes a considerable portion of the processing resources in a typical Viterbi decoder implementation. There is, therefore, a need to reduce the complexity of the history handling to permit more practical and economical implementations of a Viterbi decoder.
SUMMARY OF THE INVENTION
An object of this invention is to provide an improved method of history handling for Viterbi decoders that reduces the processing time required to decode a convolutionally encoded message relative to conventional methods.
Therefore, in accordance with an aspect of the present invention, there is provided a method of processing the state histories within a Viterbi decoder, which receives a sequence of samples, representing a sequence of encoded data bits from a convolutional encoder having N states, and a memory size of K-1 bits, where K >_ 2, N =
2K~1, and each state is uniquely identified by a sequence of K 1 state bits, which corresponds to K 1 data bits entered into the encoder. The method provided by the invention comprises the following steps. The first step is to define an initial state metric and an initial state history for each state. The second step is, for each state, insert, into the initial state history, L
contiguous state bits that correspond to the L data bits first entered into the encoder, where 1 <_ L <_ K 1. The third step is to designate the initial state metrics and the initial state histories as previous state metrics and previous state histories respectively.
The fourth step is to perform L advances, each advance using at least one branch metric determined using at least one of the samples received by the decoder, to determine N new state metrics and N new state histories from the previous state metrics and the previous state histories, where each new state history is a copy of one previous state history.
Preferably L = K-1, and L >_1. In an embodiment of the invention, at least one bit of the initial state history is present in at least one bit of the least significant bits of a storage word storing the initial state metric.
In accordance with another aspect of the present invention there is provided a state history processor within a Viterbi decoder that receives and decodes a sequence of samples representing a sequence of encoded bits obtained from encoding a sequence of data bits with a convolutional encoder having N states and a memory size of K
1 bits, where K ~, N= 2K-~, and each state is uniquely identified by a sequence of K 1 state bits corresponding to K 1 data bits entered into the encoder, the previously mentioned processing being comprised of five elements. The first element is a set of N
first buffers corresponding to the N states, where each first buffer is used for storing a previous history and a previous metric. The second element is a set of N second buffers corresponding to the N states, where each second buffer is used for storing a new history and a new metric.
The third element is a means for determining an initial state metric and an initial state history, for each state, that is to be stored in the corresponding second buffer. The fourth element is a means for inserting into each new history L contiguous state bits that correspond to L data bits first entered into the encoder, where L is at least one and at most K 1. The fifth element is a means for performing L advances, with each advance consisting of the following steps. Firstly, the first and second buffers' designations are interchanged. Secondly, at least one branch metric is determined using at least one of the samples received by the decoder. Thirdly, N new state metrics and N new state histories are determined from the previous state metrics and the previous state histories, by using the at least one branch metric, wherein each new sate history is a copy of one previous state history. Fourth, and finally, the N new state metrics and the N new state histories are stored in the N second buffers. An additional embodiment of this invention calls for a history processor, as previously described, wherein for each state at least one bit of the initial state history is present in at least one bit of the least significant bits of the initial state metric.
A significant advantage of this invention is the reduction in the computational complexity of the core of the Viterbi decoding algorithm. In the comparison of the present invention with the prior art approach of shifting each collection of history bits and then "OR-ing" in the new history bit, it becomes apparent that the present invention has a computational advantage. With the invention, there is no need either for the shift operation, or for the "OR" operation. Additionally in order to choose the "shift-and-OR"
approach to history bit handling, with embedded history, it would be necessary to ensure that the shifting did not affect the state metrics themselves. This concern is not present with this invention because the technique of embedding the history bits into the least significant bits of the state metrics works well.
Another advantage of the present invention relates to its operation on processors with word-widths less than or equal to 16 bits, also referred to as narrow word-width processors. Such processors, using conventional history handling methods, cannot feasibly embed the history bits in the least significant bits of the metric storage words, except for the case of very short constraint-lengths. This inadequacy is demonstrated by a K = 7 convolutional code (i.e. memory-6). With conventional history handling methods, the low 6 bits of history simply reproduce the state number. This means that at least 10 bits of each state metric storage word would have to be allocated for the history bits, since, for both computational and memory efficiency reasons, it is undesirable to store the history bits more often than every 4 bits processed. Thus, if 10 bits are reserved for history, only 6 bits remain for the state metrics themselves. Since the state metrics require several more bits of quantization than the input samples feeding the Viterbi decoder, the quantization of the input would be very coarse. Thus, for narrow word-width processors, using conventional history handling methods, the history bits have to be read, selected, updated, and written, independently of the state metrics. This substantially increases the processing requirements associated with the core of the Viterbi decoder. In one embodiment of the present invention, only 4 least-significant bits are needed to store the history bits, leaving 12 bits for the state metrics themselves. With the present invention, up to "memory"
history bits can be preset at once, but there is no requirement to use this maximum number. In summary, for processors with word-widths less than or equal to 16 bits, the present invention allows the use of the "embedded history" technique, which can result in a decoder throughput increase of approximately 50%.
BRIEF DESCRIPTION OF THE DRAWINGS
Exemplary embodiments of the invention will now be further described with references to the drawings, wherein:
Figure 1 illustrates a conventional rate %2 binary convolutional encoder having constraint length K = 7, as new bits enter on the right, the old bits are shifted left;
Figure 2 illustrates a binary convolutional encoder having K = 3 and generator polynomials g1 = [ 111 ]z for the top adder, and g2 = [ 1 O 1 ]2 for the bottom adder;
Figure 3 illustrates the basic structure of a Viterbi decoder associated with the encoder of Figure 2;
1 S Figure 4 illustrates the add-compare-select butterfly of a Viterbi decoder within the decoder of Figure 3;
Figure 5 illustrates, in a flow diagram, a method of Viterbi decoding where L
= K
1 preset history bits are imbedded in the least significant bits (LSB's) of each state metric, and bit decisions are output L at a time, in accordance with an embodiment of the invention;
Table 1 illustrates the conventional Viterbi Approach to storing histories whereby the LSB's of the state histories simply reproduce the state number;
Table 2 illustrates the method of pre-setting the 2 LSB's with the state at time i in accordance with an embodiment of this invention.
DESCRIPTION OF THE INVENTION
Figure 2 depicts an example of a conventional rate '/2 convolutional encoder having 2 memory units (i.e. K = 3) and 4 states (N = 2K'1). In this example, the top adder has a generator polynomial of g1 = [111]2, while the bottom adder has a generator polynomial of g2 = [101]2. Thus, differing parity bits are generated and transmitted which will allow the reconstruction of the original (raw) sequence of data bits by a Viterbi decoder. The encoder shown has 4 possible states defined by the two bits stored therein. In the state numbering convention used in this example the most significant bit is lost when a new data bit is input. This state-numbering convention is used while describing the present S invention.
Table 1 shows sets of state metrics and state histories in a conventional Viterbi decoder, while Table 2 shows sets of state metrics and state histories for an embodiment of the present invention. The state metrics are indicated as "mmm", and the state histories are shown as binary numbers.
With the conventional Viterbi decoding approach, as shown in Table 1, the least significant bits (LSB's) of the state histories simply store the same information as is inherent in the state number itself. In this case, the memory is 2, and so the two LSBs of the state histories reproduce the state number. At time i+4, the two underlined history bit positions indicate the state at time i+2 for the path that gave rise to the metric of the state being considered.
An embodiment of the invention will now be described with reference to Table 2 and with respect to a memory-2 encoder. In this embodiment the two least significant bits are preset with the state number at time i. It is to be observed that at time i+2, the two least-significant history bits associated with each state indicate the state at time i for the path that gave rise to the metric of the state being considered. As the encoder has 4 states, the Viterbi decoder also has 4 states. The 4 states are numbered according to the data bit sequence in encoding that gives rise to the state. The numbering convention is that the bit first received (oldest) in the encoder is the most significant bit of the state number, and thus the most recently received (newest) bits is the least significant bit of the state number.
The bits making up the state number are referred to as state bits. Further, the invention will be described, in particular, with reference to convolutional encoding applied in a "flushed block" manner, meaning that the encoder starts in the zero state, a finite number of data bits are fed through the encoder, and then "memory" zeros (in this case, 2 zeros) are fed into the encoder to return the encoder to the zero state.
Decoding begins by initializing the state metric for state 0 with a relatively large positive value, and initializing the state metrics for all of the other states with a relatively large negative value. The values are chosen so that state zero is guaranteed to "win" as the process of Viterbi decoding progresses. The state histories are initialized with the state numbers. This means that the state numbers are inserted into the state histories, which in this case requires 2 bits each. For example in an embodiment where the history words are each 32 bits wide, the low 2 bits of each state history indicate the state number, after initialization, while the remaining 30 bits simply hold zero bits.
When Viterbi decoding begins, the first two received samples So and S1 are used to compute two branch metrics Boo= So+S, and Bol = So-S~. Using the two computed branch metrics Boo and Bol, the Viterbi decoder can now advance to a new set of state metrics and a new set of state histories. Each new state metric and new state history is determined by two previous state metrics and corresponding state histories, as well as the two branch metrics Boo and Bol, and a knowledge of the convolutional encoder which determines which branch metric to use for each branching connection, and whether the branch metric is added or subtracted. Which two previous states are connected to a given new state is a direct consequence of the nature of a convolutional encoder. In the convolutional encoder, as each new data bit enters the encoder, the bits already in the encoder are shifted along by one bit position, and the oldest bit is discarded. With the state numbering convention chosen, for this embodiment's memory = 2 convolutional code being considered, the previous states 0 and 2 are connected to new states 0 and 1. The connections between previous states and new states is illustrated in Figure 3, which shows that each new state is reachable by 2 previous states, and itself leads to two next states.
The pattern of connections is always such that two previous states with state numbers that differ only in their most significant bit are connected to the two new states with state numbers that differ only in their least-significant bits. This pattern of two previous states connected to two new states forms the basis of the so-called "add-compare-select" butterfly operation that forms the core of the Viterbi decoding process. The connections of a single add-compare-select butterfly are shown in Figure 4, which illustrates that the state metrics and state histories of states Oxx and lxx are used in determining new state metrics and new state histories for states xx0 and xxl.
The two previous states have state numbers differing only in their most-significant bits, and the two new states have state numbers differing only in their least-significant bits.
With the present invention the manipulations of the state histories that are performed as part of the basic add-compare-select operation are eliminated.
The only history handling required of the add-compare-select butterfly operation is to select which of the two previous state histories should be copied to the new state history (which state history is copied depends on the outcome of the add-compare operation). Thus the prior art approach of equation (2), as given in the background, is replaced by equation (3) in the current invention.
H~(i) _ ~H~,(i- 1)], m = best old state Once new state metrics and new state histories have been determined, these then become previous state metrics and previous state histories for the next advance of the Viterbi decoder. Two new received samples are used to determine two new branch metrics, and add-compare-select operations are performed to determine another set of new state metrics and new state histories. This process is only repeated for a total of 2 times before some history handling operations are again performed. Two state bits were loaded initially, and so only 2 pairs of received samples are to be processed before the state bits are again inserted into the state histories.
To insert the state bits into the state histories, the state histories are first shifted left by two bits to make room for the new state bits, and then the 2 state bits forming each state number are inserted into the least-significant 2 bits of each state history.
Having inserted 2 state bits into each state history, the Viterbi decoder now advances by another set of 2 pairs of received samples, as described earlier.
This process continues in a similar manner until the entire block of received samples has been processed.
We now describe how the decoder provides output bits based on state histories.
Prior to inserting the state bits into the state histories, the decoder searches for the state metric having the largest value (i.e. the most likely state) and outputs the 2 most-significant bits of the corresponding state history. These bits will be lost in the shifting process that precedes the insertion process. At the beginning of the Viterbi decoding there are some redundant output bits, and the decoder keeps track of how many bits should be discarded. Once the entire block has been processed, the decoder outputs the bits of the state history of the known ending state (state zero), beginning with the most-significant bit. Depending on how the length of the block relates to the period-2 insertion process, there may be redundant bits output at the very end of the block.
_g_ The above describes an embodiment of the present invention in conjunction with a relatively simple Viterbi decoding process in order to illustrate the invention, which is the loading of state bits into state histories in order to eliminate all conventional history handling processes except the copying in the add-compare-select butterfly operations. In alternate embodiments, the history for each state is not stored as one long word, but portions of the state histories are stored periodically, and the output of the decoder is determined by performing a "re-trace" operation through a matrix of stored partial-history words.
Further, alternative embodiments make use of available techniques that do not require the state histories to be shifting in connection with the periodic state bit insertion operation.
Yet another embodiment makes use of an optimization technique specific to certain types of available processor platforms which is to insert the state bits into the least significant bits of the words storing the state metrics themselves. In such an embodiment, the appropriate history bits are selected automatically when the add-compare-select operation selects which state metric (plus or minus a branch metric) to store as the new state metric, thereby removing the need to separately select and copy the state history.
Figure 5 illustrates, in a flow diagram, a method of Viterbi decoding, in accordance with an embodiment of this invention, where L = K-1 preset history bits are imbedded in the least significant bits (LSB's) of each state metric, and bit decisions are output L at a time. The starting previous state metrics are initialized as described previously. The first step 10 is to insert L = K-1 preset history bits into each of the previous state metrics. This is accomplished for each of the Nprevious state metrics by inserting L = K-1 bits of a state number into the L LSB's of each corresponding state metric. The branch metrics do not use their L LSB's. The second step, 12, is to initialize a counter. The third step, 14, is to compute N new state metrics from the N previous state metrics using a portion of a Viterbi algorithm that determines the state metrics, as described previously. The history update is inherent in the selection of the winning state metric; hence, history update requires no processing. The fourth step, 16, is to designate the N new state metrics as the N previous state metrics. The fifth step, 18, is to increment the counter. The sixth step, 20, is to test the counter to see if it is equal to L. If no then repeat the previous three steps. If yes then proceed to step 22. The sixth step, 22, is to find a largest previous state metric and extract the most likely newest L history bits from the L LSB's of the metric word. The seventh step, 24, is to retrace through the stored history columns, starting with the L newest history bits, to find L most likely oldest history bits. The eighth step, 26, is to output the L oldest history bits as L bit decisions. The ninth step, 28, is to save the N previous state metrics, with L imbedded history bits, as the newest history column. This process is then repeated to obtain the next L decisions. In an alternative embodiment, an arbitrary previous state metric is used for the sixth step, 22, with a small degradation in performance.
Of course, numerous variations and adaptations may be made to the particular embodiments of the invention described above, without departing from the spirit and scope of the invention, which is defined in the claims.
FIELD OF THE INVENTION
The invention generally relates to error-correction coding or more particularly, the invention relates to history handling for Viterbi decoding.
BACKGROUND
Viterbi decoding of convolutionally encoded messages is a common error-control method in the field of digital communications systems, and is well known in the art as described in US patent 5742621, and Motorola Application Report APR40lD, revision 0, May 1998. Software based Viterbi decoders are commonly used in digital communication systems due to the availability of low cost, high performance digital signal processors (DSPs). In conventional software implementations the handling of the history, associated with the decoding path, amounts to a substantial portion of the Viterbi decoder processing time. It is common in the art to have a collection of history bits associated with each state, and a comparison/selection operation, which includes the setting of a history bit, for each state, as a method of handling the history.
Figure 1 depicts a common rate %2, constraint length K = 7, binary convolutional encoder, using a shift register with K 1 = 6 memory elements. This encoder has N = 2K-~ -26 = 64 states. For every input data bit, d(i), two coded bits, ct(i) and c2(i), are generated.
The encoder is a state machine whose state is [d(i-(K 2)),...,d(i-1),d(i)] at time i, and has generator polynomials g1 = [ 1 O 11011 ]2 and g2 = [ 1111001 ]z. The generator polynomials specify how delayed versions of the input are added up (modulo-2) to produce each output. Each binary digit (bit) of the generator polynomial corresponds to a tap on the shift register, or a tap on the input data bit. The output of the encoder is determined by the generator polynomials (as well as the state of the encoder and the input data bit), but the state transitions are independent of the polynomials, and depend only on the current state and the input data bit.
In the Viterbi algorithm (VA), both a state history Hn(i) and a state metric Mn(i) are associated with each state n, at time i. There is an associated branch metric B",n(i) for each valid transition from state m to state n. In the particular example where K =
7, states 0 (000000) and 32 ( 100000) can both go to either state 0 (000000) or state 1 (000001 ), depending on the input data bit. No other state transitions from states 0 and 32 are possible. The state metrics and the state histories are updated as follows:
Mn (1) _ ~tMm (i -1)+Bmn (t)~ ( m H" (i) _ [Hm (i -1) , LSB(n)], m = best old state (2) These two equations are referred to as the add~ompare-select (ACS) operation, and the application of the add-compare-select operation to all states at a given time is considered as one Yiterbi advance.
The application of one Viterbi advance is represented in Figure 3. In this case the memory of the convolutional encoder is 2, and hence there are 4 states. In conventional software VA implementations, two buffers are required. Each buffer must be capable of holding a full set of state metrics and histories: one for the previous, or "old" data, and one for the updated, or "new" data. After new data is obtained the new data is used as the old data for a subsequent Viterbi advance.
In determining the new state history for a state, as in equation (2), the prior art method of history handling inserts the least significant bit of the state number of the new state into the state history. This is a critical aspect of prior art history handling, where a history bit has to be inserted each time a new history is generated.
History handling consumes a considerable portion of the processing resources in a typical Viterbi decoder implementation. There is, therefore, a need to reduce the complexity of the history handling to permit more practical and economical implementations of a Viterbi decoder.
SUMMARY OF THE INVENTION
An object of this invention is to provide an improved method of history handling for Viterbi decoders that reduces the processing time required to decode a convolutionally encoded message relative to conventional methods.
Therefore, in accordance with an aspect of the present invention, there is provided a method of processing the state histories within a Viterbi decoder, which receives a sequence of samples, representing a sequence of encoded data bits from a convolutional encoder having N states, and a memory size of K-1 bits, where K >_ 2, N =
2K~1, and each state is uniquely identified by a sequence of K 1 state bits, which corresponds to K 1 data bits entered into the encoder. The method provided by the invention comprises the following steps. The first step is to define an initial state metric and an initial state history for each state. The second step is, for each state, insert, into the initial state history, L
contiguous state bits that correspond to the L data bits first entered into the encoder, where 1 <_ L <_ K 1. The third step is to designate the initial state metrics and the initial state histories as previous state metrics and previous state histories respectively.
The fourth step is to perform L advances, each advance using at least one branch metric determined using at least one of the samples received by the decoder, to determine N new state metrics and N new state histories from the previous state metrics and the previous state histories, where each new state history is a copy of one previous state history.
Preferably L = K-1, and L >_1. In an embodiment of the invention, at least one bit of the initial state history is present in at least one bit of the least significant bits of a storage word storing the initial state metric.
In accordance with another aspect of the present invention there is provided a state history processor within a Viterbi decoder that receives and decodes a sequence of samples representing a sequence of encoded bits obtained from encoding a sequence of data bits with a convolutional encoder having N states and a memory size of K
1 bits, where K ~, N= 2K-~, and each state is uniquely identified by a sequence of K 1 state bits corresponding to K 1 data bits entered into the encoder, the previously mentioned processing being comprised of five elements. The first element is a set of N
first buffers corresponding to the N states, where each first buffer is used for storing a previous history and a previous metric. The second element is a set of N second buffers corresponding to the N states, where each second buffer is used for storing a new history and a new metric.
The third element is a means for determining an initial state metric and an initial state history, for each state, that is to be stored in the corresponding second buffer. The fourth element is a means for inserting into each new history L contiguous state bits that correspond to L data bits first entered into the encoder, where L is at least one and at most K 1. The fifth element is a means for performing L advances, with each advance consisting of the following steps. Firstly, the first and second buffers' designations are interchanged. Secondly, at least one branch metric is determined using at least one of the samples received by the decoder. Thirdly, N new state metrics and N new state histories are determined from the previous state metrics and the previous state histories, by using the at least one branch metric, wherein each new sate history is a copy of one previous state history. Fourth, and finally, the N new state metrics and the N new state histories are stored in the N second buffers. An additional embodiment of this invention calls for a history processor, as previously described, wherein for each state at least one bit of the initial state history is present in at least one bit of the least significant bits of the initial state metric.
A significant advantage of this invention is the reduction in the computational complexity of the core of the Viterbi decoding algorithm. In the comparison of the present invention with the prior art approach of shifting each collection of history bits and then "OR-ing" in the new history bit, it becomes apparent that the present invention has a computational advantage. With the invention, there is no need either for the shift operation, or for the "OR" operation. Additionally in order to choose the "shift-and-OR"
approach to history bit handling, with embedded history, it would be necessary to ensure that the shifting did not affect the state metrics themselves. This concern is not present with this invention because the technique of embedding the history bits into the least significant bits of the state metrics works well.
Another advantage of the present invention relates to its operation on processors with word-widths less than or equal to 16 bits, also referred to as narrow word-width processors. Such processors, using conventional history handling methods, cannot feasibly embed the history bits in the least significant bits of the metric storage words, except for the case of very short constraint-lengths. This inadequacy is demonstrated by a K = 7 convolutional code (i.e. memory-6). With conventional history handling methods, the low 6 bits of history simply reproduce the state number. This means that at least 10 bits of each state metric storage word would have to be allocated for the history bits, since, for both computational and memory efficiency reasons, it is undesirable to store the history bits more often than every 4 bits processed. Thus, if 10 bits are reserved for history, only 6 bits remain for the state metrics themselves. Since the state metrics require several more bits of quantization than the input samples feeding the Viterbi decoder, the quantization of the input would be very coarse. Thus, for narrow word-width processors, using conventional history handling methods, the history bits have to be read, selected, updated, and written, independently of the state metrics. This substantially increases the processing requirements associated with the core of the Viterbi decoder. In one embodiment of the present invention, only 4 least-significant bits are needed to store the history bits, leaving 12 bits for the state metrics themselves. With the present invention, up to "memory"
history bits can be preset at once, but there is no requirement to use this maximum number. In summary, for processors with word-widths less than or equal to 16 bits, the present invention allows the use of the "embedded history" technique, which can result in a decoder throughput increase of approximately 50%.
BRIEF DESCRIPTION OF THE DRAWINGS
Exemplary embodiments of the invention will now be further described with references to the drawings, wherein:
Figure 1 illustrates a conventional rate %2 binary convolutional encoder having constraint length K = 7, as new bits enter on the right, the old bits are shifted left;
Figure 2 illustrates a binary convolutional encoder having K = 3 and generator polynomials g1 = [ 111 ]z for the top adder, and g2 = [ 1 O 1 ]2 for the bottom adder;
Figure 3 illustrates the basic structure of a Viterbi decoder associated with the encoder of Figure 2;
1 S Figure 4 illustrates the add-compare-select butterfly of a Viterbi decoder within the decoder of Figure 3;
Figure 5 illustrates, in a flow diagram, a method of Viterbi decoding where L
= K
1 preset history bits are imbedded in the least significant bits (LSB's) of each state metric, and bit decisions are output L at a time, in accordance with an embodiment of the invention;
Table 1 illustrates the conventional Viterbi Approach to storing histories whereby the LSB's of the state histories simply reproduce the state number;
Table 2 illustrates the method of pre-setting the 2 LSB's with the state at time i in accordance with an embodiment of this invention.
DESCRIPTION OF THE INVENTION
Figure 2 depicts an example of a conventional rate '/2 convolutional encoder having 2 memory units (i.e. K = 3) and 4 states (N = 2K'1). In this example, the top adder has a generator polynomial of g1 = [111]2, while the bottom adder has a generator polynomial of g2 = [101]2. Thus, differing parity bits are generated and transmitted which will allow the reconstruction of the original (raw) sequence of data bits by a Viterbi decoder. The encoder shown has 4 possible states defined by the two bits stored therein. In the state numbering convention used in this example the most significant bit is lost when a new data bit is input. This state-numbering convention is used while describing the present S invention.
Table 1 shows sets of state metrics and state histories in a conventional Viterbi decoder, while Table 2 shows sets of state metrics and state histories for an embodiment of the present invention. The state metrics are indicated as "mmm", and the state histories are shown as binary numbers.
With the conventional Viterbi decoding approach, as shown in Table 1, the least significant bits (LSB's) of the state histories simply store the same information as is inherent in the state number itself. In this case, the memory is 2, and so the two LSBs of the state histories reproduce the state number. At time i+4, the two underlined history bit positions indicate the state at time i+2 for the path that gave rise to the metric of the state being considered.
An embodiment of the invention will now be described with reference to Table 2 and with respect to a memory-2 encoder. In this embodiment the two least significant bits are preset with the state number at time i. It is to be observed that at time i+2, the two least-significant history bits associated with each state indicate the state at time i for the path that gave rise to the metric of the state being considered. As the encoder has 4 states, the Viterbi decoder also has 4 states. The 4 states are numbered according to the data bit sequence in encoding that gives rise to the state. The numbering convention is that the bit first received (oldest) in the encoder is the most significant bit of the state number, and thus the most recently received (newest) bits is the least significant bit of the state number.
The bits making up the state number are referred to as state bits. Further, the invention will be described, in particular, with reference to convolutional encoding applied in a "flushed block" manner, meaning that the encoder starts in the zero state, a finite number of data bits are fed through the encoder, and then "memory" zeros (in this case, 2 zeros) are fed into the encoder to return the encoder to the zero state.
Decoding begins by initializing the state metric for state 0 with a relatively large positive value, and initializing the state metrics for all of the other states with a relatively large negative value. The values are chosen so that state zero is guaranteed to "win" as the process of Viterbi decoding progresses. The state histories are initialized with the state numbers. This means that the state numbers are inserted into the state histories, which in this case requires 2 bits each. For example in an embodiment where the history words are each 32 bits wide, the low 2 bits of each state history indicate the state number, after initialization, while the remaining 30 bits simply hold zero bits.
When Viterbi decoding begins, the first two received samples So and S1 are used to compute two branch metrics Boo= So+S, and Bol = So-S~. Using the two computed branch metrics Boo and Bol, the Viterbi decoder can now advance to a new set of state metrics and a new set of state histories. Each new state metric and new state history is determined by two previous state metrics and corresponding state histories, as well as the two branch metrics Boo and Bol, and a knowledge of the convolutional encoder which determines which branch metric to use for each branching connection, and whether the branch metric is added or subtracted. Which two previous states are connected to a given new state is a direct consequence of the nature of a convolutional encoder. In the convolutional encoder, as each new data bit enters the encoder, the bits already in the encoder are shifted along by one bit position, and the oldest bit is discarded. With the state numbering convention chosen, for this embodiment's memory = 2 convolutional code being considered, the previous states 0 and 2 are connected to new states 0 and 1. The connections between previous states and new states is illustrated in Figure 3, which shows that each new state is reachable by 2 previous states, and itself leads to two next states.
The pattern of connections is always such that two previous states with state numbers that differ only in their most significant bit are connected to the two new states with state numbers that differ only in their least-significant bits. This pattern of two previous states connected to two new states forms the basis of the so-called "add-compare-select" butterfly operation that forms the core of the Viterbi decoding process. The connections of a single add-compare-select butterfly are shown in Figure 4, which illustrates that the state metrics and state histories of states Oxx and lxx are used in determining new state metrics and new state histories for states xx0 and xxl.
The two previous states have state numbers differing only in their most-significant bits, and the two new states have state numbers differing only in their least-significant bits.
With the present invention the manipulations of the state histories that are performed as part of the basic add-compare-select operation are eliminated.
The only history handling required of the add-compare-select butterfly operation is to select which of the two previous state histories should be copied to the new state history (which state history is copied depends on the outcome of the add-compare operation). Thus the prior art approach of equation (2), as given in the background, is replaced by equation (3) in the current invention.
H~(i) _ ~H~,(i- 1)], m = best old state Once new state metrics and new state histories have been determined, these then become previous state metrics and previous state histories for the next advance of the Viterbi decoder. Two new received samples are used to determine two new branch metrics, and add-compare-select operations are performed to determine another set of new state metrics and new state histories. This process is only repeated for a total of 2 times before some history handling operations are again performed. Two state bits were loaded initially, and so only 2 pairs of received samples are to be processed before the state bits are again inserted into the state histories.
To insert the state bits into the state histories, the state histories are first shifted left by two bits to make room for the new state bits, and then the 2 state bits forming each state number are inserted into the least-significant 2 bits of each state history.
Having inserted 2 state bits into each state history, the Viterbi decoder now advances by another set of 2 pairs of received samples, as described earlier.
This process continues in a similar manner until the entire block of received samples has been processed.
We now describe how the decoder provides output bits based on state histories.
Prior to inserting the state bits into the state histories, the decoder searches for the state metric having the largest value (i.e. the most likely state) and outputs the 2 most-significant bits of the corresponding state history. These bits will be lost in the shifting process that precedes the insertion process. At the beginning of the Viterbi decoding there are some redundant output bits, and the decoder keeps track of how many bits should be discarded. Once the entire block has been processed, the decoder outputs the bits of the state history of the known ending state (state zero), beginning with the most-significant bit. Depending on how the length of the block relates to the period-2 insertion process, there may be redundant bits output at the very end of the block.
_g_ The above describes an embodiment of the present invention in conjunction with a relatively simple Viterbi decoding process in order to illustrate the invention, which is the loading of state bits into state histories in order to eliminate all conventional history handling processes except the copying in the add-compare-select butterfly operations. In alternate embodiments, the history for each state is not stored as one long word, but portions of the state histories are stored periodically, and the output of the decoder is determined by performing a "re-trace" operation through a matrix of stored partial-history words.
Further, alternative embodiments make use of available techniques that do not require the state histories to be shifting in connection with the periodic state bit insertion operation.
Yet another embodiment makes use of an optimization technique specific to certain types of available processor platforms which is to insert the state bits into the least significant bits of the words storing the state metrics themselves. In such an embodiment, the appropriate history bits are selected automatically when the add-compare-select operation selects which state metric (plus or minus a branch metric) to store as the new state metric, thereby removing the need to separately select and copy the state history.
Figure 5 illustrates, in a flow diagram, a method of Viterbi decoding, in accordance with an embodiment of this invention, where L = K-1 preset history bits are imbedded in the least significant bits (LSB's) of each state metric, and bit decisions are output L at a time. The starting previous state metrics are initialized as described previously. The first step 10 is to insert L = K-1 preset history bits into each of the previous state metrics. This is accomplished for each of the Nprevious state metrics by inserting L = K-1 bits of a state number into the L LSB's of each corresponding state metric. The branch metrics do not use their L LSB's. The second step, 12, is to initialize a counter. The third step, 14, is to compute N new state metrics from the N previous state metrics using a portion of a Viterbi algorithm that determines the state metrics, as described previously. The history update is inherent in the selection of the winning state metric; hence, history update requires no processing. The fourth step, 16, is to designate the N new state metrics as the N previous state metrics. The fifth step, 18, is to increment the counter. The sixth step, 20, is to test the counter to see if it is equal to L. If no then repeat the previous three steps. If yes then proceed to step 22. The sixth step, 22, is to find a largest previous state metric and extract the most likely newest L history bits from the L LSB's of the metric word. The seventh step, 24, is to retrace through the stored history columns, starting with the L newest history bits, to find L most likely oldest history bits. The eighth step, 26, is to output the L oldest history bits as L bit decisions. The ninth step, 28, is to save the N previous state metrics, with L imbedded history bits, as the newest history column. This process is then repeated to obtain the next L decisions. In an alternative embodiment, an arbitrary previous state metric is used for the sixth step, 22, with a small degradation in performance.
Of course, numerous variations and adaptations may be made to the particular embodiments of the invention described above, without departing from the spirit and scope of the invention, which is defined in the claims.
Claims (7)
1. A method of processing state histories within a Viterbi decoder that receives and decodes a sequence of samples representing a sequence of encoded bits obtained from encoding a sequence of data bits with a convolutional encoder having N states and a memory size of K-1 bits, where:
K is at least two, N= 2k-1, and each state is uniquely identified by a sequence of K-1 state bits corresponding to K-1 data bits entered into the encoder, the method comprising steps of:
(a) for each state, determining an initial state metric and an initial state history;
(b) for each state, inserting into the initial state history L contiguous state bits that correspond to L data bits first entered into the encoder, where L is at least one and at most K-1;
(c) designating the initial state metrics and the initial state histories as previous state metrics and previous state histories, respectively; and (d) performing L advances, each advance using at least one branch metric, the branch metric determined using at least one of the samples received by the decoder, to determine N new state metrics and N new state histories from the previous state metrics and the previous state histories where each new state history is a copy of one previous state history.
K is at least two, N= 2k-1, and each state is uniquely identified by a sequence of K-1 state bits corresponding to K-1 data bits entered into the encoder, the method comprising steps of:
(a) for each state, determining an initial state metric and an initial state history;
(b) for each state, inserting into the initial state history L contiguous state bits that correspond to L data bits first entered into the encoder, where L is at least one and at most K-1;
(c) designating the initial state metrics and the initial state histories as previous state metrics and previous state histories, respectively; and (d) performing L advances, each advance using at least one branch metric, the branch metric determined using at least one of the samples received by the decoder, to determine N new state metrics and N new state histories from the previous state metrics and the previous state histories where each new state history is a copy of one previous state history.
2. The method of claim 1, wherein L = K-1.
3. The method of claim 1, wherein for each state at least one bit of the initial state history is present in at least one bit of the least significant bits of the initial state metric.
4. The method of claim 1, wherein the L advances are performed consecutively, and L >=1.
5. A state history processor within a Viterbi decoder that receives and decodes a sequence of samples representing a sequence of encoded bits obtained from encoding a sequence of data bits with a convolutional encoder having N states and a memory size of K-1 bits, where:
K is at least two, N = 2K-1, and each state is uniquely identified by a sequence of K-1 state bits corresponding to K-1 data bits entered into the encoder, the processor comprising:
(a) a set of N buffers designated as first buffers corresponding to the N
states, each first buffer for storing a previous history and a previous metric;
(b) a set of N buffers designated second buffers corresponding to the N
states, each second buffer for storing a new history and a new metric;
(c) means for determining, for each state, an initial state metric and an initial state history to be stored in a corresponding second buffer;
(d) means for inserting into each new history L contiguous state bits that correspond to L data bits first entered into the encoder, where L is at least one and at most K-1;
(e) means for performing L advances, wherein in each advance:
i) the first and second buffers' designations are interchanged;
ii) at least one branch metric is determined using at least one of the samples received by the decoder;
iii) N new state metrics and N new state histories are determined from the previous state metrics and the previous state histories, by using the at least one branch metric, wherein each new state history is a copy of one previous state history;
iv) the N new state metrics and N new state histories are stored in the N
second buffers.
K is at least two, N = 2K-1, and each state is uniquely identified by a sequence of K-1 state bits corresponding to K-1 data bits entered into the encoder, the processor comprising:
(a) a set of N buffers designated as first buffers corresponding to the N
states, each first buffer for storing a previous history and a previous metric;
(b) a set of N buffers designated second buffers corresponding to the N
states, each second buffer for storing a new history and a new metric;
(c) means for determining, for each state, an initial state metric and an initial state history to be stored in a corresponding second buffer;
(d) means for inserting into each new history L contiguous state bits that correspond to L data bits first entered into the encoder, where L is at least one and at most K-1;
(e) means for performing L advances, wherein in each advance:
i) the first and second buffers' designations are interchanged;
ii) at least one branch metric is determined using at least one of the samples received by the decoder;
iii) N new state metrics and N new state histories are determined from the previous state metrics and the previous state histories, by using the at least one branch metric, wherein each new state history is a copy of one previous state history;
iv) the N new state metrics and N new state histories are stored in the N
second buffers.
6. The state history processor of claim 5, wherein L = K-1.
7. The state history processor of claim 5, wherein for each state at least one bit of the initial state history is present in at least one bit of the least significant bits of the initial state metric.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US8759198P | 1998-06-01 | 1998-06-01 | |
| US60/087,591 | 1998-06-01 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CA2273592A1 CA2273592A1 (en) | 1999-12-01 |
| CA2273592C true CA2273592C (en) | 2006-08-22 |
Family
ID=22206104
Family Applications (3)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002273592A Expired - Fee Related CA2273592C (en) | 1998-06-01 | 1999-06-01 | Processing of state histories in viterbi decoding |
| CA002273608A Expired - Fee Related CA2273608C (en) | 1998-06-01 | 1999-06-01 | Reduced-complexity max-log app decoders and related turbo decoders |
| CA002273418A Expired - Fee Related CA2273418C (en) | 1998-06-01 | 1999-06-01 | Tail-biting turbo-code encoder and associated decoder |
Family Applications After (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002273608A Expired - Fee Related CA2273608C (en) | 1998-06-01 | 1999-06-01 | Reduced-complexity max-log app decoders and related turbo decoders |
| CA002273418A Expired - Fee Related CA2273418C (en) | 1998-06-01 | 1999-06-01 | Tail-biting turbo-code encoder and associated decoder |
Country Status (4)
| Country | Link |
|---|---|
| US (3) | US6510536B1 (en) |
| EP (2) | EP0963048A3 (en) |
| CA (3) | CA2273592C (en) |
| DE (1) | DE69936683T2 (en) |
Families Citing this family (97)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7536624B2 (en) | 2002-01-03 | 2009-05-19 | The Directv Group, Inc. | Sets of rate-compatible universal turbo codes nearly optimized over various rates and interleaver sizes |
| US6430722B1 (en) * | 1998-01-23 | 2002-08-06 | Hughes Electronics Corporation | Forward error correction scheme for data channels using universal turbo codes |
| WO2000010257A1 (en) * | 1998-08-17 | 2000-02-24 | Hughes Electronics Corporation | Turbo code interleaver with near optimal performance |
| EP2068451B1 (en) | 1998-08-27 | 2013-10-02 | Dtvg Licensing, Inc | Method for a general turbo code trellis termination |
| US6665357B1 (en) * | 1999-01-22 | 2003-12-16 | Sharp Laboratories Of America, Inc. | Soft-output turbo code decoder and optimized decoding method |
| CN100574116C (en) * | 1999-05-19 | 2009-12-23 | 三星电子株式会社 | Turbo-interleaving device and method |
| JP3846527B2 (en) * | 1999-07-21 | 2006-11-15 | 三菱電機株式会社 | Turbo code error correction decoder, turbo code error correction decoding method, turbo code decoding apparatus, and turbo code decoding system |
| GB2361855A (en) * | 1999-07-21 | 2001-10-31 | Mitsubishi Electric Corp | Turbo-code decoding using a Soft Output Viterbi Algorithm (SOVA) |
| WO2001043310A2 (en) * | 1999-12-03 | 2001-06-14 | Broadcom Corporation | Embedded training sequences for carrier acquisition and tracking |
| US6829305B2 (en) * | 1999-12-08 | 2004-12-07 | Lg Electronics Inc. | Concatenated convolutional encoder and decoder of mobile communication system |
| US6758435B2 (en) * | 1999-12-09 | 2004-07-06 | Rheinmetall W & M Gmbh | Guide assembly for a missile |
| DE10001856A1 (en) * | 2000-01-18 | 2001-08-09 | Infineon Technologies Ag | Method for decoding a data signal |
| US6898254B2 (en) * | 2000-01-31 | 2005-05-24 | Texas Instruments Incorporated | Turbo decoder stopping criterion improvement |
| US6879648B2 (en) * | 2000-01-31 | 2005-04-12 | Texas Instruments Incorporated | Turbo decoder stopping based on mean and variance of extrinsics |
| GB0004765D0 (en) * | 2000-03-01 | 2000-04-19 | Mitel Corp | Soft-decision decoding of convolutionally encoded codeword |
| US7080307B2 (en) * | 2000-03-02 | 2006-07-18 | Kawasaki Steel Corporation | Error correction decoder with correction of lowest soft decisions |
| FR2807895B1 (en) * | 2000-04-18 | 2002-06-07 | Canon Kk | ENCODING AND DECODING METHODS AND DEVICES AND SYSTEMS USING THE SAME |
| US6757859B1 (en) * | 2000-05-01 | 2004-06-29 | Zarlink Semiconductor Inc. | Parallel turbo trellis-coded modulation |
| US6732327B1 (en) * | 2000-05-05 | 2004-05-04 | Nokia Networks Oy | Scaled-feedback turbo decoder |
| US6728927B2 (en) | 2000-05-26 | 2004-04-27 | Her Majesty The Queen In Right Of Canada, As Represented By The Minister Of Industry Through The Communications Research Centre | Method and system for high-spread high-distance interleaving for turbo-codes |
| US6591395B1 (en) * | 2000-06-18 | 2003-07-08 | Silicon Integrated Systems Corporation | Memory reduction techniques in a viterbi decoder |
| FR2811169B1 (en) * | 2000-06-28 | 2002-09-06 | Canon Kk | DECODING METHOD AND DEVICE, AND SYSTEMS USING THE SAME |
| FI109162B (en) | 2000-06-30 | 2002-05-31 | Nokia Corp | Method and arrangement for decoding a convolution-coded codeword |
| US6944803B2 (en) * | 2000-07-06 | 2005-09-13 | Her Majesty The Queen In Right Of Canada, As Represented By The Minister Of Industry Through The Communications Research Centre Canada | Code structure, encoder, encoding method, and associated decoder and decoding method and iteratively decodable code structure, encoder, encoding method, and associated iterative decoder and iterative decoding method |
| JP2002064385A (en) * | 2000-08-18 | 2002-02-28 | Sony Corp | Decoder and method for decoding |
| US7127664B2 (en) | 2000-09-18 | 2006-10-24 | Lucent Technologies Inc. | Reconfigurable architecture for decoding telecommunications signals |
| US6865710B2 (en) * | 2000-09-18 | 2005-03-08 | Lucent Technologies Inc. | Butterfly processor for telecommunications |
| US6877125B2 (en) | 2000-09-18 | 2005-04-05 | Canon Kabushiki Kaisha | Devices and methods for estimating a series of symbols |
| FR2814299B1 (en) * | 2000-09-18 | 2003-01-10 | Canon Kk | METHOD AND DEVICE FOR ESTIMATING A SUCCESS OF SYMBOLS IMPLEMENTING CONDITIONAL CORRECTION, AND ASSOCIATED SYSTEMS |
| FR2814300B1 (en) * | 2000-09-18 | 2003-01-24 | Canon Kk | DEVICE AND METHOD FOR ESTIMATING A SUITE OF REDUCED IMPLANTATION SYMBOLS, CORRESPONDING CIRCUITS AND SYSTEMS |
| US7340664B2 (en) * | 2000-09-20 | 2008-03-04 | Lsi Logic Corporation | Single engine turbo decoder with single frame size buffer for interleaving/deinterleaving |
| US6604220B1 (en) * | 2000-09-28 | 2003-08-05 | Western Digital Technologies, Inc. | Disk drive comprising a multiple-input sequence detector selectively biased by bits of a decoded ECC codedword |
| JP2002111512A (en) * | 2000-09-29 | 2002-04-12 | Sony Corp | Decoding device and method, data-receiving device and method |
| US6865711B2 (en) * | 2000-12-15 | 2005-03-08 | Conexant Systems, Inc. | System of and method for decoding trellis codes |
| US7027531B2 (en) * | 2000-12-29 | 2006-04-11 | Motorola, Inc. | Method and system for initializing a training period in a turbo decoding device |
| US6952457B2 (en) * | 2000-12-29 | 2005-10-04 | Motorola, Inc. | Method and system for adapting a training period in a turbo decoding device |
| US6799295B2 (en) * | 2001-01-02 | 2004-09-28 | Icomm Technologies, Inc. | High speed turbo codes decoder for 3G using pipelined SISO log-map decoders architecture |
| US7076000B2 (en) * | 2001-01-18 | 2006-07-11 | Motorola, Inc. | Soft-decision metric generation for higher order modulation |
| FI20010147A (en) * | 2001-01-24 | 2002-07-25 | Nokia Corp | A method and arrangement for decoding a convolutionally coded codeword |
| US7003045B2 (en) * | 2001-01-31 | 2006-02-21 | Motorola, Inc. | Method and apparatus for error correction |
| US9979580B2 (en) | 2001-02-01 | 2018-05-22 | Qualcomm Incorporated | Coding scheme for a wireless communication system |
| US6961388B2 (en) * | 2001-02-01 | 2005-11-01 | Qualcomm, Incorporated | Coding scheme for a wireless communication system |
| JP3540224B2 (en) * | 2001-03-06 | 2004-07-07 | シャープ株式会社 | Turbo decoder, turbo decoding method, and storage medium storing the method |
| KR100464360B1 (en) * | 2001-03-30 | 2005-01-03 | 삼성전자주식회사 | Apparatus and method for efficiently energy distributing over packet data channel in mobile communication system for high rate packet transmission |
| JP3730885B2 (en) * | 2001-07-06 | 2006-01-05 | 株式会社日立製作所 | Error correction turbo code decoder |
| EP1292077A3 (en) * | 2001-09-11 | 2006-10-11 | Broadcom Corporation | Sequence estimation for PSK signals |
| US6760883B2 (en) * | 2001-09-13 | 2004-07-06 | Agere Systems Inc. | Generating log-likelihood values in a maximum a posteriori processor |
| US20030131306A1 (en) * | 2001-12-27 | 2003-07-10 | Yeun-Renn Ting | Method for turbo-code block message tailing and turbo-code encoder employing the same |
| US6831574B1 (en) | 2003-10-03 | 2004-12-14 | Bae Systems Information And Electronic Systems Integration Inc | Multi-turbo multi-user detector |
| US6967598B2 (en) * | 2004-02-20 | 2005-11-22 | Bae Systems Information And Electronic Systems Integration Inc | Reduced complexity multi-turbo multi-user detector |
| US6704376B2 (en) * | 2002-01-23 | 2004-03-09 | Bae Systems Information And Electronic Systems Integration Inc. | Power and confidence ordered low complexity soft turbomud with voting system |
| US7092464B2 (en) * | 2002-01-23 | 2006-08-15 | Bae Systems Information And Electronic Systems Integration Inc. | Multiuser detection with targeted error correction coding |
| WO2003088503A1 (en) * | 2002-04-05 | 2003-10-23 | Linkair Communications, Inc. | A method and device for space-time-frequency turbo encoding |
| US7107512B2 (en) * | 2002-05-31 | 2006-09-12 | Broadcom Corporation | TTCM decoder design |
| US7003703B2 (en) * | 2002-06-21 | 2006-02-21 | Sandbridge Technologies, Inc. | Method of interleaving/deinterleaving in a communication system |
| US7073114B2 (en) * | 2002-06-24 | 2006-07-04 | Massey Peter C | Method of decoding utilizing a recursive table-lookup decoding method |
| US7577207B2 (en) * | 2002-07-03 | 2009-08-18 | Dtvg Licensing, Inc. | Bit labeling for amplitude phase shift constellation used with low density parity check (LDPC) codes |
| KR100602027B1 (en) * | 2002-07-03 | 2006-07-19 | 휴우즈 일렉트로닉스 코오포레이션 | Bit-interleaved coded modulation using low density parity check ldpc codes |
| US7020829B2 (en) * | 2002-07-03 | 2006-03-28 | Hughes Electronics Corporation | Method and system for decoding low density parity check (LDPC) codes |
| US7209527B2 (en) * | 2002-07-08 | 2007-04-24 | Agere Systems Inc. | Turbo decoder employing max and max* map decoding |
| US20040019845A1 (en) * | 2002-07-26 | 2004-01-29 | Hughes Electronics | Method and system for generating low density parity check codes |
| US7864869B2 (en) * | 2002-07-26 | 2011-01-04 | Dtvg Licensing, Inc. | Satellite communication system utilizing low density parity check codes |
| JP2004080508A (en) * | 2002-08-20 | 2004-03-11 | Nec Electronics Corp | Decoding method for error correction code, its program, and its device |
| US7154965B2 (en) | 2002-10-08 | 2006-12-26 | President And Fellows Of Harvard College | Soft detection of data symbols in the presence of intersymbol interference and timing error |
| US20040092228A1 (en) * | 2002-11-07 | 2004-05-13 | Force Charles T. | Apparatus and method for enabling use of low power satellites, such as C-band, to broadcast to mobile and non-directional receivers, and signal design therefor |
| WO2004095713A2 (en) * | 2003-04-14 | 2004-11-04 | Bae Systems Information And Electronic Systems Integration Inc. | Joint symbol, amplitude, and rate estimator |
| US7246295B2 (en) * | 2003-04-14 | 2007-07-17 | Agere Systems Inc. | Turbo decoder employing simplified log-map decoding |
| US7877670B2 (en) * | 2005-12-06 | 2011-01-25 | Ternarylogic Llc | Error correcting decoding for convolutional and recursive systematic convolutional encoded sequences |
| US20110064214A1 (en) * | 2003-09-09 | 2011-03-17 | Ternarylogic Llc | Methods and Apparatus in Alternate Finite Field Based Coders and Decoders |
| US8577026B2 (en) | 2010-12-29 | 2013-11-05 | Ternarylogic Llc | Methods and apparatus in alternate finite field based coders and decoders |
| GB0323211D0 (en) * | 2003-10-03 | 2003-11-05 | Toshiba Res Europ Ltd | Signal decoding methods and apparatus |
| US20050193320A1 (en) * | 2004-02-09 | 2005-09-01 | President And Fellows Of Harvard College | Methods and apparatus for improving performance of information coding schemes |
| KR100606023B1 (en) * | 2004-05-24 | 2006-07-26 | 삼성전자주식회사 | The Apparatus of High-Speed Turbo Decoder |
| KR100549870B1 (en) * | 2004-07-09 | 2006-02-06 | 삼성전자주식회사 | Method for finding a last state in tail-biting and turbo encoder using the same |
| JP4321394B2 (en) * | 2004-07-21 | 2009-08-26 | 富士通株式会社 | Encoding device, decoding device |
| CN101341659B (en) * | 2004-08-13 | 2012-12-12 | Dtvg许可公司 | Code design and implementation improvements for low density parity check codes for multiple-input multiple-output channels |
| US7565594B2 (en) * | 2004-08-26 | 2009-07-21 | Alcatel-Lucent Usa Inc. | Method and apparatus for detecting a packet error in a wireless communications system with minimum overhead using embedded error detection capability of turbo code |
| US7971131B1 (en) * | 2005-05-06 | 2011-06-28 | Hewlett-Packard Development Company, L.P. | System and method for iterative denoising and error correction decoding |
| KR101021465B1 (en) * | 2005-10-26 | 2011-03-15 | 삼성전자주식회사 | Apparatus and method for receiving signal in a communication system using a low density parity check code |
| JP4436315B2 (en) * | 2005-12-26 | 2010-03-24 | 京セラ株式会社 | Convolutional encoder, communication apparatus, and convolutional encoding method |
| US20080063105A1 (en) * | 2006-09-13 | 2008-03-13 | Via Telecom, Inc. | System and method for implementing preamble channel in wireless communication system |
| US20080092018A1 (en) * | 2006-09-28 | 2008-04-17 | Broadcom Corporation, A California Corporation | Tail-biting turbo code for arbitrary number of information bits |
| US20080152044A1 (en) * | 2006-12-20 | 2008-06-26 | Media Tek Inc. | Veterbi decoding method for convolutionally encoded signal |
| US7992075B2 (en) * | 2007-03-19 | 2011-08-02 | Intel Corporation | Arrangements for encoding and decoding digital data |
| EP1973234A1 (en) | 2007-03-20 | 2008-09-24 | Nokia Siemens Networks Gmbh & Co. Kg | Optimised code block segmentation for turbo encoding |
| US8006172B2 (en) * | 2007-07-10 | 2011-08-23 | Oracle America, Inc. | Auxiliary path iterative decoding |
| US8375280B2 (en) * | 2007-11-02 | 2013-02-12 | Telefonaktiebolaget L M Ericsson (Publ) | Optimum distance spectrum feedforward tail-biting convolutional codes |
| US8127216B2 (en) * | 2007-11-19 | 2012-02-28 | Seagate Technology Llc | Reduced state soft output processing |
| US8250448B1 (en) * | 2008-03-26 | 2012-08-21 | Xilinx, Inc. | Method of and apparatus for implementing a decoder |
| US8630364B2 (en) * | 2008-07-25 | 2014-01-14 | Nokia Siemens Networks Oy | Termination techniques for multi-index continuous phase encoders for wireless networks |
| JP5088304B2 (en) * | 2008-11-27 | 2012-12-05 | 富士通株式会社 | Communications system |
| US20100299579A1 (en) * | 2009-05-21 | 2010-11-25 | Ternarylogic Llc | Methods and Systems for Error-Correction in Convolutional and Systematic Convolutional Decoders in Galois Configuration |
| US8683555B2 (en) * | 2010-02-10 | 2014-03-25 | Raytheon Company | Systems and methods to prevent denial of service attacks |
| US8769373B2 (en) | 2010-03-22 | 2014-07-01 | Cleon L. Rogers, JR. | Method of identifying and protecting the integrity of a set of source data |
| US9065485B1 (en) * | 2011-01-05 | 2015-06-23 | Altera Corporation | Method and apparatus for interleaving using stored initial value |
| CN103701475B (en) * | 2013-12-24 | 2017-01-25 | 北京邮电大学 | Decoding method for Turbo codes with word length of eight bits in mobile communication system |
| US11395353B2 (en) | 2020-01-23 | 2022-07-19 | Rockwell Collins, Inc. | System and method for 4G automatic link establishment protocol enhancement |
Family Cites Families (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2013002B (en) * | 1978-01-05 | 1982-03-17 | Plessey Co Ltd | Error correction devices |
| US4583078A (en) * | 1984-11-13 | 1986-04-15 | Communications Satellite Corporation | Serial Viterbi decoder |
| JPS62233933A (en) * | 1986-04-03 | 1987-10-14 | Toshiba Corp | Viterbi decoding method |
| FR2639781B1 (en) | 1988-11-25 | 1991-01-04 | Alcatel Thomson Faisceaux | INTERLEAVING METHOD FOR DIGITAL TRANSMISSION DEVICE |
| FR2675971B1 (en) | 1991-04-23 | 1993-08-06 | France Telecom | CORRECTIVE ERROR CODING METHOD WITH AT LEAST TWO SYSTEMIC CONVOLUTIVE CODES IN PARALLEL, ITERATIVE DECODING METHOD, CORRESPONDING DECODING MODULE AND DECODER. |
| US5349589A (en) | 1991-07-01 | 1994-09-20 | Ericsson Ge Mobile Communications Inc. | Generalized viterbi algorithm with tail-biting |
| US5416787A (en) * | 1991-07-30 | 1995-05-16 | Kabushiki Kaisha Toshiba | Method and apparatus for encoding and decoding convolutional codes |
| US5369671A (en) | 1992-05-20 | 1994-11-29 | Hughes Aircraft Company | System and method for decoding tail-biting code especially applicable to digital cellular base stations and mobile units |
| US5390198A (en) * | 1993-05-26 | 1995-02-14 | The Boeing Company | Soft decision viterbi decoder for M-ary convolutional codes |
| US5465275A (en) * | 1993-11-16 | 1995-11-07 | At&T Ipm Corp. | Efficient utilization of present state/next state registers |
| FR2724273B1 (en) * | 1994-09-05 | 1997-01-03 | Sgs Thomson Microelectronics | SIGNAL PROCESSING CIRCUIT FOR IMPLEMENTING A VITERBI ALGORITHM |
| DE4437984A1 (en) | 1994-10-25 | 1996-08-14 | Philips Patentverwaltung | Transmission system with soft output decoding |
| DE59509663D1 (en) * | 1995-10-21 | 2001-11-08 | Micronas Gmbh | Logical block for a Viterbi decoder |
| US6028899A (en) * | 1995-10-24 | 2000-02-22 | U.S. Philips Corporation | Soft-output decoding transmission system with reduced memory requirement |
| US5742621A (en) | 1995-11-02 | 1998-04-21 | Motorola Inc. | Method for implementing an add-compare-select butterfly operation in a data processing system and instruction therefor |
| JP3233847B2 (en) * | 1996-02-23 | 2001-12-04 | 沖電気工業株式会社 | Viterbi decoding method and Viterbi decoding circuit |
| US5764649A (en) | 1996-03-29 | 1998-06-09 | Amati Communications Corporation | Efficient address generation for convolutional interleaving using a minimal amount of memory |
| US6023783A (en) * | 1996-05-15 | 2000-02-08 | California Institute Of Technology | Hybrid concatenated codes and iterative decoding |
| US5719875A (en) | 1996-06-11 | 1998-02-17 | Lucent Technologies Inc. | Systematic convolution interleavers and deinterleavers |
| US5734962A (en) * | 1996-07-17 | 1998-03-31 | General Electric Company | Satellite communications system utilizing parallel concatenated coding |
| GB2317789A (en) * | 1996-09-26 | 1998-04-01 | Lin Mao Chao | Decoding trellis code using Viterbi algorithm |
| US5933462A (en) * | 1996-11-06 | 1999-08-03 | Qualcomm Incorporated | Soft decision output decoder for decoding convolutionally encoded codewords |
| KR19990012821A (en) * | 1997-07-31 | 1999-02-25 | 홍성용 | Electromagnetic wave absorber composition and its manufacturing method, electromagnetic wave absorbing coating composition, its manufacturing method and its coating method |
| EP0897223B1 (en) * | 1997-08-14 | 2013-03-20 | Her Majesty The Queen In Right Of Canada as represented by the Minister of Industry | High-performance low-complexity error-correcting codes |
| JP4033245B2 (en) * | 1997-09-02 | 2008-01-16 | ソニー株式会社 | Turbo coding apparatus and turbo coding method |
| US6275538B1 (en) * | 1998-03-11 | 2001-08-14 | Ericsson Inc. | Technique for finding a starting state for a convolutional feedback encoder |
| US6192501B1 (en) * | 1998-08-20 | 2001-02-20 | General Electric Company | High data rate maximum a posteriori decoder for segmented trellis code words |
| US6014411A (en) * | 1998-10-29 | 2000-01-11 | The Aerospace Corporation | Repetitive turbo coding communication method |
| US6044116A (en) * | 1998-10-29 | 2000-03-28 | The Aerospace Corporation | Error-floor mitigated and repetitive turbo coding communication system |
| US6266795B1 (en) * | 1999-05-28 | 2001-07-24 | Lucent Technologies Inc. | Turbo code termination |
-
1999
- 1999-06-01 US US09/324,429 patent/US6510536B1/en not_active Expired - Fee Related
- 1999-06-01 US US09/324,812 patent/US6530059B1/en not_active Expired - Fee Related
- 1999-06-01 US US09/324,428 patent/US6460161B1/en not_active Expired - Lifetime
- 1999-06-01 EP EP99304246A patent/EP0963048A3/en not_active Withdrawn
- 1999-06-01 CA CA002273592A patent/CA2273592C/en not_active Expired - Fee Related
- 1999-06-01 DE DE69936683T patent/DE69936683T2/en not_active Expired - Fee Related
- 1999-06-01 CA CA002273608A patent/CA2273608C/en not_active Expired - Fee Related
- 1999-06-01 CA CA002273418A patent/CA2273418C/en not_active Expired - Fee Related
- 1999-06-01 EP EP99304227A patent/EP0963049B1/en not_active Expired - Lifetime
Also Published As
| Publication number | Publication date |
|---|---|
| EP0963049B1 (en) | 2007-08-01 |
| CA2273608C (en) | 2007-08-07 |
| DE69936683D1 (en) | 2007-09-13 |
| EP0963048A2 (en) | 1999-12-08 |
| US6510536B1 (en) | 2003-01-21 |
| CA2273608A1 (en) | 1999-12-01 |
| CA2273418C (en) | 2008-03-25 |
| CA2273418A1 (en) | 1999-12-01 |
| US6530059B1 (en) | 2003-03-04 |
| EP0963049A2 (en) | 1999-12-08 |
| EP0963048A3 (en) | 2001-02-07 |
| CA2273592A1 (en) | 1999-12-01 |
| EP0963049A3 (en) | 2001-03-28 |
| US6460161B1 (en) | 2002-10-01 |
| DE69936683T2 (en) | 2008-04-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA2273592C (en) | Processing of state histories in viterbi decoding | |
| US5355376A (en) | Circular viterbi decoder | |
| KR100538730B1 (en) | Viterbi decoding apparatus and viterbi decoding method | |
| AU724573B2 (en) | Convolutional decoding with the ending state decided by CRC bits placed inside multiple coding bursts | |
| US5436918A (en) | Convolutional encoding/decoding apparatus with fixed bit insertion | |
| US5802116A (en) | Soft decision Viterbi decoding with large constraint lengths | |
| CN101431388B (en) | Method and apparatus of decoding encoded data frame | |
| US6272661B1 (en) | Minimum memory implementation of high speed viterbi decoder | |
| EP0355850B1 (en) | Sequential decoder | |
| US4715037A (en) | Viterbi decoder comprising a majority circuit in producing a decoded signal | |
| GB2315001A (en) | Viterbi decoder for depunctured codes | |
| KR100336246B1 (en) | Integrated circuit with digital processor and co-processor | |
| US20070201586A1 (en) | Multi-rate viterbi decoder | |
| KR100262303B1 (en) | Survivor path trace back method in decoding with viterbi algorithm and apparatus thereof | |
| EP1024603A2 (en) | Method and apparatus to increase the speed of Viterbi decoding | |
| FI100564B (en) | A method for generating transition metrics and a receiver for a cellular radio system | |
| US6578119B2 (en) | Method and device for memory management in digital data transfer | |
| JP3272173B2 (en) | Error correction code / decoding device | |
| JPH06121172A (en) | Picture encoder | |
| JP3235333B2 (en) | Viterbi decoding method and Viterbi decoding device | |
| US5666380A (en) | Digital communication system and method of compressing information | |
| JP3351732B2 (en) | Arithmetic encoding apparatus and method, arithmetic decoding apparatus and method, and recording medium | |
| KR100399410B1 (en) | Viterbi decoder and decoding method thereof | |
| WO2010064496A1 (en) | Decoding method and decoding device | |
| JP2004193892A (en) | Viterbi decoder |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| EEER | Examination request | ||
| MKLA | Lapsed |