34results about How to "Reduce decoding overhead" patented technology

An apparatus for performing LSTM neural network operations and an operating method are provided. The device comprises a direct memory access unit, an instruction cache unit, a controller unit, a plurality of data cache units arranged in parallel and a plurality of data processing modules arranged in parallel, wherein the plurality of data processing modules are in one-to-one correspondence with the data cache units, and are used for obtaining input data from the corresponding data cache units and weights and offsets required during operation, and performing LSTM neural network operation; and parallel operation is executed among the plurality of data processing modules. According to the method, special instructions are adopted for operation, the number of instructions required by operationis greatly reduced, and decoding expenditure is reduced; caching the weight and the bias, so that the data transmission overhead is reduced; the method is not limited to the specific application field, can be used in the fields of voice recognition, text translation, music synthesis and the like, and is high in expandability; and the plurality of data processing modules operate in parallel, so that the operation speed of the LSTM network is remarkably improved.

The invention provides an LT code degree distribution design method and device. A bit error rate and the decoding cost can be reduced. The decoding efficiency is improved. The method comprises the steps of determining a packet number of a to-be-coded source data packet based on a distribution function for PMRSD degree distribution; generating a coding generation matrix G <KxN > according to the total number K of source data packets, the total number N of coding data packets and the packet number of the to-be-coded source data packet; carrying out module 2 multiplication on the source data packets and the G <KxN >, thereby obtaining the coding data packets, and sending the coding data packets to a receiving end; generating the coding generation matrix G <KxN > at the receiving end, judging whether a list of which degree value d=1 exists in the generated G <KxN > or not, if the list exists, assuming that the source data packet s corresponding to a point (i, j) of the degree value d=1 in the G <KxN > is equal to the received coding data packet c<j >, and adding up a decoding counter count and 1; and assuming that a value of the coding data packet c<j > related to the s is equal to the exclusive or of the value of the original coding data packet c<j > and the s, setting a value of an element related to the s in the G <KxN > as 0, and finishing decoding when the count is equal to K. The method and the device relate to the technical field of communication.

The present invention relates to a method for encoding and decoding an online fountain code with unequal error protection strategy, comprising: step A: an encoder selects a window W according to an extended window strategy 1 or W 2 , the coded symbol with coding degree 2 is sent to the receiving end; Step B: the receiving end receives the coded symbol with coding degree 2 and the corresponding source code symbol sequence number and is used to form different connection groups; Step C: Repeat step A and step B, When the number of elements in the connection group with the largest number of elements in MISs reaches the preset value k 1 alpha 0 When the receiving end feeds back to the encoding end, the encoding end stops encoding and sending the coded symbols with the coding degree of 2 after receiving the feedback; Step D: the coding end obtains the coded symbols with the coding degree of 1 from the MISs and sends it to the receiving end; Step E : The receiving end receives the coded symbol with a coding degree of 1 and the corresponding source symbol serial number and uses it to decode the connection group, etc. Compared with the prior art, the present invention has the advantages of reducing decoding overhead on the premise of ensuring unequal error protection performance.

The invention relates to a fountain-code encoding method with feedback and the unequal error protection capacity. Under some application scenarios, parts of data sets need to be encoded preferentially. The fountain-code encoding method includes the steps of firstly, sequentially processing all input symbols through Bernoulli random experiments; secondly, starting to encode the input symbols through an encoder with an encoding method of standard LT codes; thirdly, when the serial number of a codeword received by a decoder is larger than or equal to the number of the input symbols, feeding the number of most important bits recovered through the decoder and the number of least important bits recovered through the decoder back to the encoder; fourthly, designing a most important increasing-degree generator according to the feedback information to replace a standard-degree generator to encode the most important bits, and designing a least important increasing-degree generator according to the feedback information to replace a standard-degree generator to encode the least important bits. According to the fountain-code encoding method, an encoding structure of the standard LT codes is improved, different encoding methods are adopted for the bits with different important degrees, and the important bits are preferentially recovered.

The invention provides a construction method of degree distribution of LT codes. The method first adjusts a binary system degree distribution, then organically combines the binary system degree distribution with robust soliton distribution, and further optimizes a degree distribution function through optimization of a translatable set value, thereby obtaining degree distribution which also has relatively good performance when code length of source data is short, i.e., binary system-robust soliton distribution is corrected. The corrected binary system-robust soliton distribution is utilizes to perform LT fountain coding on the source data, decoding spending can be reduced, decoding efficiency can be improved, and fountain codes can be well applied to the communication fields.

The invention discloses a degree distribution function design method of network layer fountain code and channel joint encoding. The method comprises the following steps: importing a channel deletion probability and an error correction probability of MP joint decoding into an EISD degree distribution function, and determining the degree distribution function of a signal source to send an encoding packet by an expression of a channel deletion and error correction probability influence function and the expression of a Per-Pc-EISD degree distribution function; importing a strong robust factor, and determining an expanded Per-Pc-EISD degree distribution function based on the channel deletion probability and the error correction probability; performing transfer and normalization processing on the Per-Pc-EISD degree distribution function, and determining the degree distribution function of joint encoding of network layer fountain code encoding and physical layer channel encoding. According to the degree distribution function design method disclosed by the invention, when the Per-Pc-EISD degree distribution function is designed, the probability of restoring a plurality of wrong data packets during the encoding and the channel deletion probability are imported into the degree distribution function, with respect to the joint encoding method of adopting fountain code encoding on a network layer and adopting channel encoding on a physical layer, the decoding cost is effectively reduced, and the reliability and speed of wireless transmission are improved.

The invention provides a digital fountain code encoding and decoding method based on ARM. The invention relates to a digital fountain code encoding and decoding method. The problems of repeated selection, large decoding cost, large number of iterations and complicated calculation of the existing digital fountain code LRLTC algorithm are solved. The method comprises the steps that two positive integers are generated randomly and uniformly; sum (j) is acquired according to a robust solitary wave distribution formula; a parameter is randomly selected between 0 and 1, so that sum(j)<r<sum(j+1) and d=j; if X and d satisfy the condition that 1<X<K-1 and d is equal to or greater than 3, the period length in which d is repeated is T; if d is equal to or less than T, d is not repeatedly selected; each original data packet participating in encoding is numbered; according to d, the numbered original data packet is selected for encoding; and the encoded original data packet is transmitted to a decoding module for decoding, so as to realize encoding and decoding. The method provided by the invention is used in the field of channel encoding.

The invention relates to a high-coding symbol value broadcasting method based on system online fountain codes, and belongs to the technical field of channel coding, fountain codes and wireless broadcasting. The method comprises the following steps: 1) performing initialization; 2) generating and sending a first type of coding symbols; 3) receiving and processing the first type of coding symbols; 4) repeating the steps 2) and 3) until the sending number of the first type of coding symbols reaches a set parameter; 5) enabling a decoder to count the recovery proportion and generate feedback; 6) generating and sending a second type of coding symbols; 7) receiving and processing the second type of coding symbols; 8) processing the coded symbol with the degree value of 1; 9) processing the codedsymbol with the degree value of 2; 10), repeating steps 5) -9) till decoders at all receiving ends finish coding. under the condition that broadcast channel receiving ends are heterogeneous and are not provided with the decoders, the number of feedback retransmission times during fountain code decoding is effectively reduced, the decoding expenditure of users with good channel states and poor channel states is reduced, and the system throughput is improved.

The invention discloses a signal processing method and device, comprising: obtaining the position of expected continuous error bits in the data block to be transmitted; obtaining the position of the target interleaving bit in the data block to be transmitted; wherein, the target interleaving bit is obtained by less than the preset The number of iterations of the iteration threshold can recover the bits of the information; the bits at positions where continuous error bits are expected to occur and the bits at positions where target interleaved bits are expected to be interleaved with each other. It can be seen from the embodiment of the present invention that since the target interleaved bits are interleaved to the position where the continuous error bit is expected to occur, the expected continuous error bit is interleaved to the position where the information can be easily recovered, so that the position where the continuous error bit is expected to occur Both are bits that are easy to recover information, thus simplifying the subsequent decoding process and reducing the decoding overhead.

The invention relates to a low-delay channel coding method combining LT codes and multiple connections, and belongs to the technical field of channel coding. The method optimizes the degree distribution function of the LT code, that is, determines the degree distribution function of the joint degree distribution of the dual proportional coefficients, and determines the optimal proportional coefficient according to the channel conditions, and uses the optimized degree distribution function to generate coded symbols, through Two links are used for transmission, and the encoded symbols received in the two links are combined at the receiving end to perform confident hard-decision decoding. This method achieves better performance than HARQ and traditional multi-connection in terms of decoding delay and communication system stability indicators. The LT code after optimizing the degree distribution function has achieved a significant reduction in bit error rate during transmission. Overhead can be lower than 10 at 0.5 ‑5 .