Low complexity compression of short messages

1. A for data communication, which method comprises at least a first data compression performed before transmission, which method further comprises at least a first data decompression, which is performed after the data is received, the data compression and decompression being performed by use of a specific hash table, by which data compression a symbol for each data segment is generated, which symbols are transmitted to the receiver, in which receiver the symbols are decompressed by use of the specific hash table, whereby characterized in that the specific hash table is performed as a hash table array, which hash table array generates symbols comprising at least a hash count and a parity check, in which hash table array the parity check comprises generating of a data string, which data string is compared with a parity stored in the hash table, where if both parities are the same, the data model returns the count in the hash table element for transmission, and where, if a zero count is returned indicating that the data string is not valid in the hash table, no transmission is performed.

2. The method according to claim 1 , whereby characterized in that the method concerns a specific hash function, comprising a data array in which each element containing two entries, which method further concerns a parity check function, in which parity check function, the string is given to a One-at-a-Time hash function, which function transfers the string of data characters to an integer number within an interval, which interval contains a first tablesize, which first tablesize denotes the total number of elements in the hash table.

3. The method according to claim 2 , whereby the required memory and the complexity is reduced by using a context model, which context model for the first comprises a data model and for the second functions for estimating statistics, which statistics comprises at least a total, a left, and a right count, which counts are generated on the fly, which data model at least comprises the symbol counts.

4. The method according to claim 3 , whereby each single symbol counts refers to a single node in a data three.

5. The method according to claim 4 , whereby the data model is parameterized by just one parameter, which parameter is the size of the hash table, where each count of the hash table refers to an arbitrarily long string, whereby the model works for each statistical context model order without any changes in the implementation.

6. The method according to claim 5 , whereby a statistical model consists of an array of elements, where each of the array elements contains the count and the parity, which numbers are calculated by a more complex statistical context model for adaptive modeling, where the data model consists of a hash table with collision resolving techniques, where training of the context model is performed by the text file, from which text file the training data shall be extracted for compression, while the more complex hash table with linked elements is filled with statistical data.

7. The method according to claim 6 , whereby the compression is finished, the amount of statistical data in the data model is reduced, where each linked list is searched for the maximum statistical item count, and the maximum counts are written into the first elements of the linked list, where the remaining elements of the linked lists are deleted.

8. The method according to claim 7 , whereby the appending strings to the counts of the table are recalled for to calculate the parities, where a parity of one table element is calculated by a logical exclusive-OR operation of the single symbols of one string, for obtaining a low complexity data model, which data model is an array of elements, where each element containing a count and a parity.

9. The method according to claim 8 , whereby a low-complexity data model consists of a subspace of the original data model, which contains the context nodes that are of statistical importance.

10. The method according to claim 9 , whereby an indicator to assess the statistical quality of the retained nodes are the frequency each node is accessed during the compression.

11. The method according to claim 9 , whereby lossless text compression is performed by algorithms comprising a statistical context model, which model comprises an arithmetic coder and a context model library, which context model uses a data structure with linked lists for resolving collisions, where each node of a data tree is represented by a list item, which contains a pointer to a key and selected node statistics comprising a symbol and a total count, where a bit-mask is employed to indicate present successors of the node.