Patent Application
20110103519
The present invention relates to systems and methods for decoding data, and more particularly, to systems and methods for decoding information additive codes and multi-stage information additive codes, herein referred to collectively as “chain reaction codes.”
Chain reaction codes have been described previously in the assignee's patents, such as U.S. Pat. No. 6,307,487 entitled “Information Additive Code Generator and Decoder for Communication Systems” (hereinafter “Luby I”), and U.S. patent application Ser. No. 10/032,156, entitled “Multi-Stage Code Generator and Decoder for Communication Systems” (hereinafter “Raptor”). As described therein, chain reaction decoding is a unique form of forward error-correction that enables data reconstruction from a received data set of a given size, without regard to the particular data packets received. Communication systems employing chain reaction codes are able to communicate information much more efficiently compared to traditional FEC codes transmitted via data carousel or acknowledgement-based protocols, as described in Luby I or Raptor.
Once the source symbols have been created, the output symbols are generated from the source symbols. Hereinafter, an output symbol and an input symbol are described as “associated” if the value of the input symbol is used to obtain the value of the output symbol. The mathematical operation which defines this association may be any particular operation, and in one embodiment, the output symbol's value is the XOR of the values of some of the source symbols. For each output symbol, key generator 140 produces a key, from which the weight of the output symbol is determined from a weight table 150. Once the weight W is determined, W random or pseudorandom source symbols are chosen, and the value of the output symbol is computed as the XOR of the values of these source symbols. For example, in
In a typical chain reaction decoding process, decoding starts by identifying an output symbol O1 associated with a single source symbol. The term “output symbol of degree one” refers to the aforementioned output symbol associated with only one source symbol. Similarly, an output symbol associated with two source symbols would be referred to as an output symbol of “degree two.” Source symbols are referred to in a similar manner corresponding to the number of output symbols each source symbol is associate with.
Once the output symbol O1 of degree one is identified, the associated source symbol of O1 is recovered and is removed from the decoding graph. The process continues by identifying another output symbol O2 of degree one. For example, in the situation depicted in
The process is continued until all the source symbols are recovered, or until there is no output symbol of degree one. For example, in the situation of
In this case decoding is successful.
The foregoing chain reaction decoding process encounters difficulty when no output symbol of degree one is found. In some instances, the decoding process may stop prematurely and the decoder may flag an error. Alternatively, the decoder may use other more elaborate algorithms like Gaussian elimination to complete decoding, if possible. However, the running time of Gaussian elimination may be prohibitively large for applications where fast decoding is desired, especially when the number of unrecovered input symbols at the time when no more output symbols of degree one are found is large. This would lead to a decoding algorithm whose computational overhead is substantially larger than a chain reaction decoder, and may therefore be undesirable in certain applications.
For this reason, the design of chain reaction coding systems usually is done in such a way to guarantee that the decoder does not stop prematurely. This requirement may put stringent conditions on the design of the chain reaction code than may be possible using a more complex decoder. For example, it may enforce the average degree of an output symbol to be higher than otherwise, and thus may lead to a decrease in the performance of the encoder and of the decoder. More generally, this decoding procedure forces the design of the weight table to be in such a way as to guarantee the success of the abovementioned decoding algorithm with high probability, and hence may put restrictions on the set of possible weight tables.
What is therefore needed is a new decoding algorithm that offers similar computational advantages as the chain reaction decoder, and is able to continue decoding even if no output symbol of degree one is found at some stage of the decoding.
The present invention provides systems and processes for decoding a chain reaction code, even when no output symbol of degree one is found in the code. This is accomplished in one embodiment by selecting a source symbol which is associated to an output symbol of degree two or higher (i.e., an output symbol which is itself associated with two or more input symbols). The source symbol associated with the output symbol of degree two or higher is then deactivated in an attempt to produce an output symbol of degree one. The inactivation process can be repeated either successively until an output symbol of degree one is identified, and/or whenever the decoding process is unable to locate an output symbol of degree one. Various embodiments of the processes and systems are presented herein.
For clarity and convenience, features and components which are identified in earlier drawings retain their reference numerals in subsequent drawings.
The following terms are used throughout the application and are intended to have the indicated meaning:
The term “active” refers to a possible state of a source symbol. The active state of a source symbol is not permanent, and the active state of a source symbol may change to either an “inactive” state, a “recoverable state”, or a “recovered” state as these terms are defined below.
The terms “deactivated” or “inactive” refers to another state of a source symbol. The state of a deactivated source symbol is not necessarily permanent, and an inactive source symbol may be reactivated in processes under the present invention.
The term “recoverable” refers to yet another state of a source symbol indicating that the value of the source symbol can be recovered if the values of some other source symbols are recovered. In a particular embodiment of the invention, a source symbol may become “recoverable” through the inactivation of one or more source symbols.
The term “recovered source symbol” refers to a source symbol whose values has been determined. The value of a source symbol may be determined either directly, e.g., from the value of an output symbol to which is singly associated therewith, or indirectly, e.g., from the value of a deactivated source symbol.
In one embodiment of the invention, recovery of source symbols described in 310 occur temporally before the recovery of deactivated and recoverable source symbols referred to in 320. However, the invention is not limited thereto, and recovery of the source symbols identified in 310 may occur substantially concurrently with the recovery of the deactivated and recoverable source symbols in process 330 in alternative embodiments of the present invention.
If the deactivation process of 322 does result in the production of one or more output symbols of degree one, the process continues at 324 where the source symbol associated with an output symbol of degree one is declared recoverable. The process then returns to 323 where a determination is made as to whether any additional output symbols of degree one remain. The processes of 323 and 324 are repeated until all of the output symbols of degree one produced by the preceding deactivation process are declared recoverable.
If the deactivation of the selected source symbol at 322 does not result in an output symbol of degree one, or once all of the source symbols associated with an output symbol of degree one are declared recoverable at 324, the process continues from 323 to 325, where a determination is made as to whether any source symbols associated with output symbols of degree two or higher remain. If so, the process returns to 321 where another active source symbol associated with an output symbol of degree two or higher is selected, deactivated, and the presence of output symbols of degree one is checked. One or more iterations of the processes may occur, for instance, where the deactivation of a first source symbol associated with an output symbol of degree two or higher does not result in an output symbol of degree one, but additional source symbols associated with an output symbol of degree two (or higher) remain. In this case, the subsequent deactivation of another source symbol associated with an output symbol of degree two (or higher) may produce one or more output symbols of degree one. The process repeats until all source symbols have been either been recovered (via the start-up process 310), deactivated (via 322), or declared recoverable (via 325), at which point the process proceeds to the source symbol value recovery process 330.
Those of skill in the art will appreciate that other selection criteria may be used to select source symbols in order to obtain the largest number of output symbols of degree one. For example in one process, the source symbol associated with the largest number of output symbols is selected for deactivation. In another embodiment, a source symbol is randomly selected from a group of those source symbols associated with two or more output symbols. In still a further embodiment, an output symbol is identified which is associated with a predetermined number of source symbols, e.g., the fewest. Subsequently, all but one of the source symbols is selected for deactivation.
In another embodiment of the source symbol selection process, a string of source symbols may be recovered. In this process, an output symbol of degree two is to be identified such that one of its associated source symbols is itself associated with a second output symbol of degree two, and such that one of its associated source symbols is itself associated with a third output symbol of degree two, and so on. Such a chain of output symbols will be called a degree-two-chain hereinafter.
Any of the source symbol selection processes may further include a “back-tracking process” by which the deactivated source symbol is reactivated, and another source symbol is selected for deactivation in accordance with the methods presented herein. The invention is not limited to the exemplary processes by which a source symbol is selected for deactivation, and any method in which a source symbol associated with two or more output symbols is selected can be used in the present invention.
As explained above with reference to
P·M·Q·(Q−1·x)=P·b,
where x is the column vector (x1, . . . , xK), and b is the vector consisting of the values of the K′ received output symbols. In practice, the matrices P and Q may not be stored as full matrices, but as permutations computed by tracking the process of the Inactivation Decoding. This form usually requires much less memory than the storage of a complete matrix. As can be appreciated by those skilled in the art, the recovery process does not depend on the specific permutation of the columns of the illustrated decoding matrix, and other column permutations may be used in alternative embodiments under the present invention.
Of the many ways possible for computing the solution x of the system of equations given above, we will illustrate in the following one possibility. This is served for descriptive purposes only and is not intended to limit the scope of this invention.
For the description of the core of the algorithm, it is advantageous to denote the vector Q−1·x by y, and redefine the task of decoding as the task of computing the vector y. Once y is computed, x may be efficiently computed as the permutation of y described by Q. Further, the matrix P·M·Q is denoted by N; the vector P·b is denoted by c, that is, c is the permutation of b described by P, which is again efficient to compute. The task is then to calculate the vector y satisfying N·y=c, where N has the shape given in
To solve this system, Gaussian elimination may be applied to matrix N. The rows of the submatrix B are eliminated by the rows of the lower triangular matrix L. The same transformation is applied to the vector c. This action transforms the matrix B into the matrix consisting of zeros, and the matrix C is transformed into a different matrix D, obtained by applying the same elimination steps to the matrices A and C. This transformation is shown in
This elimination transforms the original system of equations into two separate systems: the system given by D·z=e, and the system L·u+A·z=d. The values of the unknown vector z correspond to the values of the source symbols corresponding to the inactivated source symbols. Once these values are found from the set of equations D z=e, the remaining values given by u can be found in a variety of ways. In some embodiments of the present invention, these values can be found by multiplying the matrix A with z, XOR'ing the resulting vector with d to obtain a vector g, and solving the system of equations L·u=g. In some embodiments, the latter system may be solved using a chain reaction decoder. In yet other embodiments, the value of each source symbol corresponding to an inactive source symbol is XOR'd with the values of the output symbols corresponding to the neighboring output symbols associated to said source symbol, and the inactive source symbol is removed from the corresponding decoding graph (not shown). This produces a new restricted decoding graph with all the inactive source symbols removed. Then a normal chain reaction decoding may be applied to the restricted Decoding Graph to recover the other source symbols.
The system of equations D·z=e can be solved in a variety of ways. In some embodiments, this system may be solved using the Gaussian elimination algorithm. In other embodiments, the inactivation decoding may be applied recursively to obtain the unknown values of the inactive source symbols. Other methods for solving systems of linear equations may also be applied.
In some embodiments of the inactivation decoder, the decoding process may begin before all the output symbols have been entered into the decoding graph. In these embodiments, whenever the decoding graph has no more output symbols of degree one and has at least one active source symbol, the above-described strategies may be employed to determine whether to inactivate a source symbol or whether to enter another output symbol into the Decoding Graph if such an output symbol exists. In cases where the decoding process begins before all the output symbols have been collected, the creation of the decoding matrix, and the elimination process for the decoding matrix may happen substantially concurrently with the reception process, with one or more steps of the elimination process being done with the reception of every new output symbol. Alternatively, more than one output symbol could be collected at a time, and decoding could proceed until all the said output symbols are processed; if not all source symbols are recovered at this point, another set of output symbols could be requested and processed, until all the source symbols have been recovered.
As a particular example, where a low-density parity-check code is used for the static encoding process, then a number of multi-stage output symbols equal to the number of check symbols in this code may be added to the decoding graph, their value set to 0, and the decoding graph may be augmented by the graph of the low-density parity-check code between the source symbols and the check symbols, and the decoding graph may be replaced by the new graph. The choice of low-density parity-check codes is not essential to this application. In general, for any type of static encoding, the corresponding parity-check matrix defines a bipartite graph by which the decoding graph may be augmented.
In some embodiments of this invention, inactivation decoding may proceed in the same manner as described above, with the decoding graph of
Each of the methods described herein may be practiced in a multitude of different ways (i.e., software, hardware, or a combination of both) and in a variety of systems. In one embodiment, the described methods can be implemented as instruction codes stored either on a computer readable disk, in memory (volatile or non-volatile), or reside within a processor (computer, embedded processor, and the like). In addition, a system for decoding a chain reaction code using the inactivation techniques described herein may comprise a computer or other such programmable machine having a memory operable to store and/or execute instruction codes corresponding to the processes described herein.
Accordingly, in some embodiments of the present invention the inactivation decoding mechanism is used to reduce the reception overhead of chain reaction coding when the entire original content needs to be reconstructed.
In other embodiments of the present invention, the inactivation decoder is used to reduce the average degree of an output symbol, and hence decrease the computational resources used for creating output symbols.
Another property of a chain reaction coding system using an inactivation decoder is that a weight table can be designed in which none of the output symbols may be of degree one. This means that none of the output symbols of such a coding system contains the value of an input symbol. In some embodiments, this property can be used to reduce the average degree of the output symbols, thereby decreasing the computational load on the encoder. Moreover, in some applications, this property may be used to give the transmission a light level of security against unauthorized access to the original data.
Inactivation Decoding Example
An embodiment of a chain reaction coding system as disclosed in Raptor is described by the number of data symbols, denoted N, a static encoding which generates R static encoding symbols, and a dynamic encoder described by a weight table. In some embodiments a reception overhead may also be specified that gives good probabilistic guarantees of the success of the decoder. In other embodiments, output symbols may be collected until complete decoding is possible, and there is no need for specifying a reception overhead.
The following table describes various parameters for an exemplary inactivation decoder, with the first column giving the range for the value N, the second giving information on the generation of static encoding symbols, the third giving the weight table for the generation of dynamic encoding symbols, and finally the fourth giving the number of static encoding symbols computed:
For example, the ninth row in the table means that if N is between 5000-8100, then the number R of static encoding symbols is the smallest integer greater than or equal to 0.05*N+100. In all the cases, the first stage of the static encoder may use first a Hamming code to encode the original symbols, as described in Raptor. The second stage may use a low-density parity-check code. In the example given by the ninth row, the parity-check matrix of this code consists of two submatrices. The first has └2*R/3┘ rows and N+R columns, where └a┘ denotes the largest integer smaller than or equal to a. The second submatrix has R−└2*R/3┘ rows and N+R columns. Each of these submatrices is picked randomly subject to the condition that in the first matrix each column has exactly 1 nonzero entry, and in the second matrix each column has exactly 7 nonzero entries.
The weight tables corresponding to S0, S1, and S2 are given by:
The average weights of S1 and S2 are 4.254 and 4.154, respectively.
If the reception overhead is fixed at 5% or 50, whichever is larger, then it can be mathematically proven that the probability of failure of the inactivation decoding is less than 10−13. The concrete example given here is only for illustrative purposes. Variations of the actual numbers lead to designs which are within the scope of this invention.
As can be seen, in some of the runs the number of inactive source symbols is zero, meaning that the normal chain reaction decoder would have completed the decoding. However, where N is between 1 and 10,000, in the majority of the cases the number of inactive source symbols is larger than one. In these cases the normal chain reaction decoder would have failed. The number of inactivated source symbols is very often zero if the number N of source symbols is larger than 20,000. In these cases, the decoder is particularly fast, while giving exceptionally good probabilistic guarantees on successful decoding.
The foregoing description has been presented for purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed, and obviously many modifications and variations are possible in light of the above teaching. The described embodiments were chosen in order to best explain the principles of the invention and its practical application to thereby enable others skilled in the art to best utilize the invention in various embodiments and with various modifications as are suited to the particular use contemplated. It is intended that the scope of the invention be defined by the claims appended hereto.
The following references are herein incorporated by reference in their entirety for all purposes:
The present application claims the benefit of U.S. Provisional Application No. 61/235,285, filed Aug. 19, 2009, entitled “Methods and Apparatus Employing FEC Codes With Permanent Inactivation of Symbols for Encoding and Decoding Processes,” and assigned to the assignee hereof and herein incorporated by reference for all purposes. The present application is a continuation-in-part of, and claims priority to, U.S. patent application Ser. No. 11/842,102, filed Aug. 20, 2007, now U.S. Pat. No. 7,633,413, issued Dec. 15, 2009, which is a continuation of U.S. patent application Ser. No. 11/356,303, filed Feb. 15, 2006, now U.S. Pat. No. 7,265,688, issued Sep. 4, 2007, which is a continuation of U.S. patent application Ser. No. 11/031,331, filed Jan. 7, 2005, now U.S. Pat. No. 7,030,785, issued Apr. 18, 2006, which is a continuation of U.S. patent application Ser. No. 10/459,370, filed Jun. 10, 2003, now U.S. Pat. No. 6,856,263, issued Feb. 15, 2005, which claims the benefit of U.S. Provisional Application No. 60/388,129, filed Jun. 11, 2002, the contents of each of which are herein incorporated by reference in their entirety for all purposes.
| Number | Date | Country | |
|---|---|---|---|
| 61235285 | Aug 2009 | US | |
| 60388129 | Jun 2002 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 11356303 | Feb 2006 | US |
| Child | 11842102 | US | |
| Parent | 11031331 | Jan 2005 | US |
| Child | 11356303 | US | |
| Parent | 10459370 | Jun 2003 | US |
| Child | 11031331 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 11842102 | Aug 2007 | US |
| Child | 12549560 | US |
