CN103124181A - Turbo code decoding iteration cease method based on cosine similarity - Google Patents

Abstract

The invention discloses a turbo code decoding iterative stop method based on cosine similarity, which includes the following steps: (1) perform an iterative decoding between two component decoders; (2) store the output of the component decoder II The external information sequence of the block to be decoded is defined as an external information vector; (3) Calculate the cosine similarity between the current iteration and the external information vector of the previous iteration; (4) The cosine similarity calculated in step (3) If it is greater than the threshold or the current iteration number reaches the preset maximum value, then enter step (5); otherwise, repeat steps (1)-(4); (5) translate the last iteration component The logarithmic likelihood ratio generated by encoder II is deinterleaved and hard-decided, and is output as the decoding result of the decoder. The method significantly improves the iterative decoding speed when decoding the Turbo code pattern in the LTE/LTE-A system, and has certain value for designing a high-rate Turbo decoder meeting the requirements of the LTE/LTE-A system.

Description

An Iterative Stopping Method for Turbo Code Decoding Based on Cosine Similarity

technical field

The present invention relates to LTE/LTE-A technology, relates to Turbo code high-speed decoding technology, in particular to a turbo code decoding iteration stop method based on cosine similarity.

Background technique

Turbo code is a channel coding technique with superior performance. The performance of Turbo codes is improved by exchanging external information between two component decoders through mutual iterative processes. The more iterations, the better the decoding performance, but at the same time, the higher the complexity, the greater the delay. Modern communication systems require higher and higher transmission rates. How to effectively reduce the decoding delay of Turbo codes and how to balance decoding performance and delay are very important research topics.

Both the channel coding in the 3GPP Long Term Evolution LTE system and the LTE-A system use Turbo codes. The LTE-A system standard requires a downlink peak rate of 1Gbit/s and an uplink peak rate of 500Mbit/s. In order to meet the high transmission rate requirements of LTE-A system 1Gbit/s, it is necessary to improve the traditional turbo codec, design and verify a new decoding algorithm architecture that meets the high transmission rate requirements of future wireless communication systems. The iterative stopping strategy is one of the key technologies to improve the decoding speed, which can greatly reduce the average number of iterations in the turbo decoding process and improve the decoding speed with little bit error rate performance loss.

The bit error performance of the turbo code decreases continuously with the iteration, but after a certain number of iterations, the performance of the turbo code does not improve with the iteration, and the continuous iteration at this time will only increase the system delay in vain. Moreover, for some data sequences, error-free decoding can be achieved after a small number of iterations; for some data sequences, due to the existence of too many errors, it is impossible to completely correct errors no matter how many iterations are performed. Therefore, it is unnecessary to set the same fixed number of times for all sequences to be decoded in the traditional method, which will cause waste of system resources and time; and dynamically determine the decoding iterations of each data sequence according to a certain iteration stop strategy The number of times can effectively reduce the number of decoding iterations and improve the average decoding speed of the Turbo decoder under the condition that the performance of the Turbo code is not greatly affected.

The criteria used to measure different iterative stopping methods mainly include decoding speed (the average number of iterations per decoding frame), bit error performance (bit error rate and frame error rate), and the complexity of the stopping criterion. The iterative stopping method needs to greatly reduce the average number of iterations in the process of turbo decoding with little bit error performance loss, improve the decoding speed, and consider the complexity of the algorithm at the same time.

Some commonly used iterative stopping methods such as hard decision aided (HDA) criterion, improved hard decision aided criterion (IHDA) criterion and sign difference ratio (SDR) criterion have low complexity. The HDA criterion stops iterating when the hard decision symbol of the soft information output by the two consecutive iterations of the component decoder II no longer changes; the IHDA criterion improves the HDA criterion, and the purpose is to reduce the storage of the last iteration information, thereby reducing Storage requirement; SDR criterion determines whether to continue iteration by comparing the prior information of component decoder I or component decoder II with the different bit number and threshold of external information symbols in the same iteration. These criteria are based on the sign of soft information in the iterative process as a measure, and the average number of iterations is far from the ideal criterion. The cyclic redundancy check (CRC) criterion performs a CRC check on the hard decision result, and if the result is 0, the iteration is stopped; the CRC criterion has a higher speed, but a higher complexity.

Contents of the invention

The purpose of the present invention is to provide a turbo code decoding iteration stop method based on cosine similarity, which can greatly reduce the average number of iterations in the turbo decoding process and further improve the decoding speed under the condition of small bit error rate performance loss.

In order to achieve the above object, the technical solution of the present invention is:

A turbo code decoding iterative stop method based on cosine similarity, comprising the following steps: (1) perform an iterative decoding between component decoder I and component decoder II of the turbo decoder; (2) store The external information of each bit of the block to be decoded outputted by the component decoder II in step (1), the external information of the first bit to the last bit in the block to be decoded outputted by the component decoder II The information sequence is regarded as a vector, called the external information vector, and the modulus length of the external information vector is calculated and stored; (3) From the second iteration, the external information vector obtained in step (2) is calculated and the external information vector of the previous iteration Cosine similarity of the information vector; (4) Compare the cosine similarity calculated in step (3) with the threshold corresponding to the preset current iteration number, if the cosine similarity is greater than the threshold or the current iteration number reaches the preset the maximum number of iterations, then go to step (5); otherwise, repeat steps (1), (2), (3), (4); (5) for the log likelihood generated by component decoder II in the last iteration Deinterleaving and hard decision are performed on the ratio, and the hard decision is output as the final decoding result of the Turbo code decoder.

The specific method of step (1) in the present invention is: (11) before the first iteration, the prior information of the component decoder I is initialized to 0; (12) the system information bit, the check bit of the component decoder I and the prior information are input to the component decoder I for MAP decoding to generate the extrinsic information and log likelihood ratio of the component decoder I; (13) The extrinsic information of the component decoder I after QPP interleaving is used as the component The prior information of the decoder II, the interleaved system information bits and the parity bits of the component decoder II are input to the component decoder II for MAP decoding, and the external information and logarithm of the component decoder II are generated. Likelihood ratio; (14) The external information of component decoder II after QPP deinterleaving will be used as the new prior information of component decoder I.

The external information vector at the nth iteration in step (2) of the present invention is a K-dimensional vector composed of the external information sequence from the first bit to the last bit in the block output by the component decoder II, K The block size for the message.

The modulus length of the extrinsic information vector at the nth iteration in the step (2) of the present invention is the arithmetic square root of the sum of the squares of the extrinsic information of all bits in the block output by the component decoder II;

In the step (3) of the present invention, the cosine similarity of the external information vector of the nth iteration and the n-1th iteration is used as a measure, wherein the cosine similarity of the two vectors is defined as:

和

为两个维度相同的矢量，

和

分别为矢量

和的模长，A_i和B_i分别为

和的第i个分量。In the formula,

and

are two vectors of the same dimension,

and

are vectors

and The modulus length of A _i and _Bi are respectively

and The i-th component of .

When calculating the cosine similarity in step (3) of the present invention, the magnitude of the absolute value of the external information is 10-100. In order to avoid overflow due to excessive intermediate values in the calculation process, all external information can be divided by 100 to participate in the calculation , the size of the final cosine similarity remains unchanged;

A set of threshold values corresponding to the current iteration number n in step (4) of the present invention, that is, the threshold at the nth iteration is θ _n , n=2,3,...,N _max -1; where N _max is the preset maximum number of iterations. The range of θ _n is θ _n ∈ [0.85,1], and θ _n-1 ≤ _{θ n} ; the specific size of θ _n needs to be determined according _to the simulation; High and Low are set to two or more different values.

In yet another embodiment of the present invention, the cosine similarity calculated in step (3) is compared with the corresponding threshold θ _n , and if it is greater than the threshold, the iterative process is stopped, that is, the iteration stop condition is:

cosα＞ _θn

Wherein, α is the angle between two external information vectors, and the number of iterations n satisfies 2≤n≤N _max -1.

In yet another embodiment of the present invention, a maximum number of iterations N _max is preset, and when the number of iterations reaches N _max , the iterative process is stopped no matter whether the iteration stop condition is met or not.

In yet another embodiment of the present invention, the two information vectors in the cosine similarity are the external information of the component decoder, the prior information or the information sequence in the decoding process such as the logarithmic likelihood ratio or transform it The subsequent information sequence;

Compared with the prior art, the present invention uses the cosine similarity between external information vectors in the iterative decoding process as a measure, further improving the decoding speed. The present invention further reduces the average number of iterations in the turbo decoding process and improves the decoding speed under the condition that the bit error rate performance loss is small.

Description of drawings

FIG. 1 is a flow chart of the iterative stopping method of Turbo code decoding based on cosine similarity in the present invention.

FIG. 2 is a decoding structure diagram of the turbo code decoding iteration stop method based on cosine similarity shown in FIG. 1 .

Detailed ways

The present invention will become clearer through the following description in conjunction with the accompanying drawings, which are used to explain the embodiments of the present invention.

Embodiments of the present invention will now be described with reference to the drawings, in which like reference numerals represent like elements.

The turbo code decoding iterative stop method based on cosine similarity in the present invention is realized based on the Turbo code pattern in LTE/LTE-A; the channel type is an additive white Gaussian noise channel (AWGN); the modulation mode adopts a binary phase shift key control (BPSK); the decoding algorithm uses MAP decoding; the Turbo code generator polynomial is (13,15); the code rate is 1/3.

The flow of the iterative stop method for decoding Turbo codes based on cosine similarity in this embodiment will be described in detail below. In conjunction with Fig. 1, Fig. 2, then described method comprises the following steps:

Step S1, initialize the prior information of the component decoder I to 0, and the cosine similarity to 0;

Step S2, comparing and judging whether the cosine similarity between the nth iteration and the n-1th iteration external information vector is greater than the corresponding preset threshold θ _n , or whether the number of iterations n is greater than the maximum number of iterations N _max ; 2≤n≤ N _max -1, if at least one condition is established, go to step S8; otherwise, go to the next step;

Step S3, the number of iterations n is increased by 1;

Step S4, the external information of the last iteration of the component decoder II is deinterleaved by QPP, and used as the prior information of the component decoder I (the prior information of the component decoder I is the initialization value 0 in the first iteration), and System information bits and check bits of component decoder I are input to component decoder I for MAP decoding to generate external information and log likelihood ratio of component decoder I;

Step S5, the external information of the component decoder I after QPP interleaving is used as the prior information of the component decoder II, and the interleaved system information bits and the check bits of the component decoder II are input to the component decoder Ⅱ performs MAP decoding to generate external information and log likelihood ratio of component decoder Ⅱ;

Step S6, store the external information vector output by the component decoder II, calculate and store the modulus length of the external information vector, the external information vector is the first bit to the last bit in the block output by the component decoder II The K-dimensional vector composed of the external information sequence, K is the size of the information block, and the modulus length of the external information vector is the arithmetic square root of the sum of the squares of the external information of all bits in the block output by the component decoder II;

Step S7, calculate the cosine similarity of the external information vector of the nth iteration and the n-1th iteration; the cosine similarity of the two vectors is defined as:

和

为两个维度相同的矢量，

和分别为矢量和的模长，A_i和B_i分别为

和

的第i个分量。返回至步骤S2；In the formula,

and

are two vectors of the same dimension,

and are vectors and The modulus length of A _i and _Bi are respectively

and

The i-th component of . Return to step S2;

Step S8, stop the iterative process, perform deinterleaving and hard decision on the log-likelihood ratio generated by the component decoder II last time, and obtain the final decoding result; end.

It can be seen from the above that the cosine similarity-based Turbo code decoding iteration stop method in this embodiment uses the cosine similarity of the external information of the two component decoders in the iterative decoding process to measure the continuous passing of the two decoders. The ability to improve the decoding performance by exchanging external information, so as to judge whether it is enough to stop the iterative process to improve the decoding speed. The invention further reduces the average number of iterations in the turbo decoding process and improves the decoding speed under the condition that the bit error rate performance loss is small.

The present invention has been described above in conjunction with the best embodiments, but the present invention is not limited to the above-disclosed embodiments, but should cover various modifications and equivalent combinations made according to the essence of the present invention.

Claims (8)

1. A Turbo code decoding iterative stop method based on cosine similarity, is characterized in that comprising the steps: (1) Carry out an iterative decoding between the component decoder I and the component decoder II of the Turbo decoder ; (2) Store the external information of each bit of the block to be decoded output by the component decoder II in step (1), and store the first bit to the last bit in the block to be decoded output by the component decoder II The external information sequence of one bit is regarded as a vector, called the external information vector, and the modulus length of the external information vector is calculated and stored; (3) From the second iteration, the external information vector obtained in step (2) and The cosine similarity of the external information vector of the last iteration; (4) compare the cosine similarity calculated in step (3) with the preset threshold corresponding to the current iteration number, if the cosine similarity is greater than the threshold or the current iteration If the number of iterations reaches the preset maximum number of iterations, go to step (5); otherwise, repeat steps (1), (2), (3), (4); (5) generate the component decoder II in the last iteration The logarithm likelihood ratio is used for deinterleaving and hard decision, and this hard decision is output as the final decoding result of the Turbo code decoder. 2. The iterative stop method of Turbo code decoding based on cosine similarity according to claim 1, characterized in that: the specific method of the step (1) is: (11) before the first iteration, component decoder I (12) System information bits, parity bits of component decoder I and prior information are input to component decoder I for MAP decoding to generate external information of component decoder I and Logarithmic likelihood ratio; (13) The external information of the component decoder I after QPP interleaving is used as the prior information of the component decoder II, and the interleaved system information bits and the parity bits of the component decoder II Input to component decoder II for MAP decoding to generate external information and log likelihood ratio of component decoder II; (14) The external information of component decoder II after QPP deinterleaving will be used as component decoder Encoder I new prior information is used. 3. The turbo code decoding iterative stop method based on cosine similarity according to claim 1, characterized in that: the extrinsic information vector during the nth iteration in the step (2) is the component decoder II output A K-dimensional vector composed of the external information sequence from the first bit to the last bit in the block, where K is the size of the information block. 4. The iterative stopping method of Turbo code decoding based on cosine similarity according to claim 3, characterized in that: the modulus length of the external information vector during the nth iteration in the step (2) is component decoder The arithmetic square root of the sum of the squares of the external information of all bits in the block output by II. 5. The iterative stop method of Turbo code decoding based on cosine similarity according to claim 1, characterized in that: the step (3) adopts the difference between the external information vector of the nth iteration and the n-1th iteration Cosine similarity is used as a measure, where the cosine similarity of two vectors is defined as:

In the formula,

and

are two vectors of the same dimension, and

are vectors

and

The modulus length of A _i and _Bi are respectively

and

The i-th component of . 6. The iterative stopping method of Turbo code decoding based on cosine similarity according to claim 5, characterized in that: when calculating the cosine similarity in the step (3), the magnitude of the absolute value of the external information is 10-100, In order to avoid overflow caused by excessive intermediate values in the calculation process, all external information is divided by 100 to participate in the calculation, and the final cosine similarity remains unchanged. 7. the Turbo code decoding iterative stop method based on cosine similarity according to claim 1, is characterized in that: two information vectors in the described cosine similarity are external information, prior information of component decoder Or the information sequence in the decoding process such as the logarithmic likelihood ratio or the information sequence after it is transformed. 8. The iterative stopping method of Turbo code decoding based on cosine similarity according to claim 1, characterized in that: a set of threshold values corresponding to the current number of iterations n in the step (4), that is, the nth The threshold for the first iteration is θ _n , n=2,3,…,N _max -1; where N _max is the preset maximum number of iterations; the range of θ _n is θ _n ∈ [0.85,1], and θ _{n -1} ≤ θ _n .

Images