KR100850744B1 - LLR computing device and method - Google Patents

Abstract

The present invention relates to an apparatus and method effective for calculating LLRs using block combining with an improved MAP algorithm.

An LLR calculation method embodying the idea of the present invention includes the steps of: calculating an alpha value, a beta value, and a gamma value in at least two time intervals; Obtaining a transition probability value in the at least two time intervals of each state; A comparison operation is performed on some of the transition probability values to determine a large value, a value is selected from the remaining ones of the transition probability values according to the determined value, and a large value is compared with the determined value. Obtaining a maximum value of the transition probability values in a manner of selecting; And determining the operation to be applied according to the maximum value to obtain the LLR.

LLR, Transition Probability, Viterbi Decoder

Description

LLR computing device and method

1 is a trellis diagram from time k-1 to k + 1 in a 4-state prior art.

2 is a trellis diagram applying an improved prior art 4-state block processing technique from time k-1 to k + 1.

3 is a block integrated trellis diagram from time k-1 to k + 1 of 4-state determined by input data of two time intervals, in accordance with the present invention;

4 is a flowchart illustrating a process of obtaining a maximum value of a transition probability value according to the present invention.

5 is a block diagram showing an LLR calculation apparatus according to an embodiment of the present invention.

FIG. 6 is a circuit diagram illustrating an embodiment of the transition probability value calculator of FIG. 5. FIG.

The present invention relates to an effective method and apparatus for calculating an LLR using block combining, which is an improvement of a maximum a posteriori (MAP) algorithm.

With the development of communication systems, a fast and reliable channel coding scheme is required for information transmission. However, information loss occurs due to noise, fading, interference, etc. of the communication channel. It is essential to use error correction code to minimize information loss and correct errors. The study of error corrections has been studied in many places since the publication of Shannon's findings in 1948, among which turbocodes were proposed by Berrou, Glavieux, and Thitimaishima in 1993, with very good error correction capabilities that approach the limits of Shannon's method. The research on this is being actively conducted. The decoding of the turbo code restores original information through iterative decoding using a MAP (Maximum a Posteriori) decoder or a SOVA (Soft-Output Viterbi Algorithm) decoder. MAP algorithm has more complexity than SOVA algorithm, but has superior performance in terms of performance. Therefore, MAP algorithm with excellent BER performance is frequently used.

The MAP algorithm was first proposed in 1974 by Bahl et al. And is an algorithm for calculating APP from a noisy signal. The purpose of the MAP decoding algorithm is to determine the most likely information bit for the received symbol after the data is received. This is called the Log Likelihood Ratio (LLR). Figure 1 shows a trellis diagram from time k-1 to k + 1 of a 4-state.

Here alpha is a forward state metric (FSM), which is a state for transition from the previous state S 'of time k-1 to the next state S for time k with the information bits. Represents a metric. Beta (β) is called a backward state metric (BSM). Like alpha, after receiving all information, the current beta value may be repeatedly obtained from the beta value of the previous state. Gamma (γ) is defined as branch metric (BM).

In order to calculate the LLR, the gamma value is obtained from the received data, and the alpha and beta values are calculated using the obtained gamma value. Equations applied to this process are shown in Equations 1 and 2.

In order to calculate the LLR, the gamma value is calculated from the received data, and the alpha and beta values are calculated using the gamma value. The formula is as follows.

Where u _k is the data bit and v _k is the parity bit. x _k is the noisy data bits that pass through the channel, and y _k is the noisy parity bits that pass through the channel. Once you have the alpha, beta, and gamma values, you can use these values to calculate the LLR as

Equations 4 and 5 are formulas for calculating LLRs at times k and k + 1, respectively. Meaning of the above equations is to determine the information bits with the highest probability when coming from the previous state to the current state.

However, the MAP algorithm of the above-described method requires a large memory size and a complicated computational amount, which increases the design burden of the system and requires a high cost of implementation.

In order to improve the disadvantage of the MAP algorithm that requires a large memory size as described above, a block processing algorithm has been proposed. That is, the block processing algorithm is a MAP algorithm that can use memory more efficiently. The principle is as follows.

FIG. 2 is a trellis diagram applying a block processing technique from time k−1 to k + 1 of a 4-state. Algorithm using block processing method calculates alpha value and beta value from time k-1 to time k + 1 at k + 1 instead of time k. This has the advantage of reducing memory to store and store beta values. The equation for calculating the alpha value and the beta value is shown in Equations 6 to 8 below.

However, in the algorithm using the block processing technique, the memory consumption and power are reduced by reducing the number of data accesses by using the efficient memory in the conventional MAP algorithm, but the multiplication operation is increased when comparing the LAP calculation with the conventional MAP algorithm. The problem is that the decoding speed drops.

The present invention has been made to solve the above problems, and an object of the present invention is to provide an LLR calculation method and apparatus that can improve decoding speed while improving memory use efficiency in turbo decoding.

In addition, another object of the present invention is to provide a method and apparatus for calculating an LLR that can simplify the hardware configuration and maintain decoding efficiency.

To this end, an object of the present invention is to provide an LLR calculation method and apparatus capable of minimizing multiplication and comparison operations.

In order to achieve the above object, the LLR calculation method of the present invention uses the block processing described above, but instead of using the complex formula used in the conventional improvement technique to obtain the LLR, any of the state transition probability values of each state is the largest. As a result, there is an improvement in selecting and using one of a plurality of LLR calculation formulas.

That is, a maximum value of the transition probability values of each state is obtained, and a large value is determined by performing a comparison operation on some of the transition probability values, and one value from the other of the transition probability values is determined according to the determined value. The main idea is to select a large value by selecting and comparing the determined value with the selected value.

As described above, the present invention provides a method for calculating an LLR using a block processing technique, the method comprising: obtaining an alpha value, a beta value, and a gamma value in two or more time intervals; Obtaining a transition probability value in the at least two time intervals of each state; A comparison operation is performed on some of the transition probability values to determine a large value, a value is selected from the remaining ones of the transition probability values according to the determined value, and a large value is compared by comparing the determined value with the selected value. Obtaining a maximum value of the transition probability values in a manner of selecting; And determining the operation to be applied according to the maximum value to obtain the LLR.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. This will be described in detail by calculating the LLR for the 4-state for the transition in the 2-time interval.

In the LLR calculation method of this embodiment, first, an alpha value (FSM), a beta value (BSM), and a gamma value (BM) for each state in two time intervals (k-1 to k + 1) are obtained.

In this embodiment, a block consolidation technique is used to reduce multiplication in the MAP algorithm to which the block processing technique is applied. FIG. 3 is a trellis diagram from time k-1 to k + 1 of the 4-state. 3 illustrates a process of integrating two decoding processes into one according to input data of a two time interval and decoding the same. The alpha (FSM) and beta (BSM) values are calculated using the combined calculated gamma (BM) values for each state. The meaning of the combined gamma (BM) value for each state is that the gamma (BM) value from time k-1 to time k and the gamma (BM) value from time k to k + 1 are calculated. At this time, the equation used is the same as Equation 9 to Equation 11 below.

The value of γ _k ^{s ', s "} means that the state S' at time k-1 transitions from the state S of time k to the next state S" from the state S of time k. It is the integrated calculated value. The integrated gamma value is used to find the alpha and beta values of time k + 1. Here, the alpha and beta values are the same as the calculation method of the existing MAP algorithm. That is, alpha represents a state metric that will transition to any state of time k + 1 in all states of time k-1. Beta is also obtained repeatedly from the previous beta value.

Alternatively, it may be obtained using Equations 6 to 8 in the same manner as in the block processing algorithm described above as an improved prior art.

Once the alpha, beta, and gamma values of each state have been obtained, the transition probability values in the two time intervals of each state are obtained using the obtained values.

When transitioning to the '0' state of time k + 1 in all four states of time k-1, the data is received (0,0), and the transition to the '1' state of time k + 1 This is the case when data (0, 1) is received. When the transition to the '2' state at time k + 1 is performed, data is received (1,0). When the transition to the '3' state at time k + 1 is performed, data is transferred to (1,1). Is received. The probability value for the case where the transition to each state is calculated as shown in Equations 12 to 15 below.

p (u _k = 00: y _k ) represents the probability that the data value received when receiving data y _k is (0,0), and p (u _k = 01: y _k ) represents the data received when receiving data y _k Denotes the probability that is (0,1). Similarly, p (u _k = 11: y _k ) represents the probability of (1,0), and p (u _k = 11: y _k ) represents the probability of (1,1).

Once the probability values for each are found, the maximum of them is determined. The reason for determining the maximum value is due to the calculation process that depends on the maximum value when calculating the LLR by determining the path when the transition from time k-1 to k + 1. The process of obtaining the maximum value can be simplified by using the characteristics of the turbo decoding algorithm. Accordingly, the method of obtaining the simplified maximum value is as follows.

FIG. 4 is a schematic flowchart illustrating a process of obtaining a maximum probability value of the present embodiment for determining the largest value of the transition probability values. As shown in the drawing, the step of obtaining the maximum probability value may be divided into the following detailed processes.

First step: comparing a transition probability value for (0,0) with a transition probability value for (0,1) and selecting a large value.

Second Step: If the transition probability value for (0,0) is greater in the first process, select a transition probability value for (1,0), and if the transition probability value for (0,1) is larger, (1 Selecting a transition probability value for 1).

The third step: selecting a larger value between the selection value of the first process and the selection value of the second process as a maximum value.

In the first process, a large value is selected by comparing A1 and A2 and stored in the first register. If A1 is selected in the first step, A3 is stored in the second register. On the other hand, if A2 is selected in the first process, A4 is stored in the second register. Finally, a large value is determined and output by comparing the stored values of the first register and the second register in the third process.

6 illustrates a circuit structure for performing the process of obtaining the maximum probability value. In FIG. 6, the first register serves as the selection of the first MUX 242, and the second register serves as the switching of the switch 244.

A1 (0,0), A2 (0,1), A3 (1,0), A4 (1,1) of the data of FIGS. 5 and 6 A1 (0,0), A2 (0,1) MUX1 242 is compared to select one of the two. If MUX1 242 has selected A1 (0,0), the control signal causes switch 244 to automatically select A3 (1,0). Otherwise, if MUX1 242 selects A2, the control signal automatically selects A4 (1,1). The MUX2 246 selects the larger value of the value selected by the MUX1 242 and the two values selected from the control signal and outputs the output.

A1 (0,0) is a value obtained by adding a probability that the first decoded data is '0' and a probability that the second decoded data is '0'. That is, the probability that L (u _k ) is '0' and the probability value that L (u _{k + 1} ) is '0' are added. A2 (0,1) is the probability that the first decoded data is '0', the second decoded data is '1', and A3 (1,0) is the first data is '1'. The probability that the first data is '0' is added. A4 (1,1) is also the probability value of the probability that the first data is '1' and the second data is '1'.

When A1 (0,0) is selected in MUX1 242, the reason why the control signal should select A3 (1,0) is that A1 (0,0) means that L (u _k ) is' 0. The probability that '(') is equal to the probability that L (u _{k + 1} ) is '0' plus the probability that L (u _k ) is '0' and L (u _{k + 1} ) is '1' Because it is bigger. That is, since the probability that L (u _k ) is '0' is equal and the probability that L (u _{k + 1} ) is '0' is greater than the probability value that is '1', L (u _{k + 1} ) is '1'. The comparison of probability values is excluded in the next step, and only the probability values of L (u _k ) equal to '0' and '1' are compared. Therefore, if a1 (0,0) is selected, the control signal automatically selects A3 (1,0) because only the probability value of L (u _k ) is '1' and the probability value of '0'. Control switch 244.

Similarly, if A2 (0,1) is selected in MUX1 242, the control signal selects a4 (1,1). This means that a2 (0,1) adds the probability that L (u _k ) is '0' and the probability that L (u _{k + 1} ) is '1', and that L (u _k ) is '0', L The probability that (u _{k + 1} ) is '0' is greater than the sum. That is, since the probability that L (u _k ) is '0' is equal and the probability that L (u _{k + 1} ) is '1' is greater than the probability value that is '0', L (u _{k + 1} ) is '0' The comparison of one probability value is excluded in the next step, and only the probability value that L (u _k ) is '0' and the probability value that is '1' are compared. Therefore, if A2 (0,1) is selected, the control signal automatically selects A4 (1,1) because only the probability value of L (u _k ) is '1' and the probability value is '0'. Done.

When the maximum probability value is determined by the above process, the LLR is obtained by applying different formulas according to the determined maximum probability value. The LLR calculation formula applied according to each maximum probability value is represented by the following Equations 16 to 19 and same.

L (u _k ) is the decoded value at time k and L (u _{k + 1} ) is the decoded value for time k + 1. Dividing p (u _k = 10: y _k ) by p (u _k = 00: y _k ) in L (u _k ) of case1 means that p (u _k = 00: y _k ) is k at time k-1 Since the probability value that receives 0 and transitions to k + 1 at time k is multiplied by the probability value that receives 0 and transitions at time k, the value decoded at time k + 1 must be offset. As shown in Equation 16. To this end, _{p (u k = 10: y} k) a p: was divided by _{(u k = 00 y k)} .

By doing this we can get the decoded value for L (u _k ). L (u _{k + 1)} similarly to offset the probability to go to k at time k-1 can be obtained by the L (u _{k + 1)} with zoom. In case2, case3 and case4, L (u _k ) and L (u _{k + 1} ) are calculated.

Meanwhile, Equation 12 may be further simplified as shown in Equation 20 below.

Since the alpha value of time k + 1 includes alpha values for all states of the previous state, the equation can be simplified only by the alpha and beta of the current state. have. Similarly, equations (13) to (15) can be simplified as well.

An embodiment of an LLR calculation apparatus for performing the above-described LLR calculation method is illustrated in FIG. 5. The illustrated LLR calculator includes a forward state metric (FSM) calculator 140 for obtaining an alpha value in two or more time intervals; A reverse state metric (BSM) calculator 160 for obtaining beta values in at least two time intervals; A branch metric (BM) calculation unit 120 for obtaining a gamma value in two or more time intervals; A transition probability value calculating unit (220) for obtaining a transition probability value in the two or more time intervals of each state from the alpha value, the beta value, and the gamma value; A maximum probability value determiner 240 for obtaining a maximum value of the transition probability values; And an LLR calculator 260 for calculating an LLR by performing an operation specified according to the maximum probability value with respect to the transition probability values.

The branch metric calculator 120 calculates a gamma value of two or more time intervals using Equation 9 by the block processing algorithm.

The forward state metric calculation unit 140 receives the gamma value obtained by the branch metric calculation unit 120 and calculates an alpha value in two or more time intervals by applying Equation 11 using the block processing algorithm. .

The reverse state metric calculation unit 160 receives a gamma value obtained by the branch metric calculation unit 120 and calculates a beta value in two or more time intervals by applying Equation 10 using the block processing algorithm. .

The transition probability calculator 220 receives the gamma, alpha, and beta values obtained by the branch metric calculator 120, the forward state metric calculator 140, and the reverse state metric calculator 160. The transition probability values are obtained by applying the equations (12) to (15).

An embodiment of the maximum probability value determiner 240 used in the LLR calculating apparatus of FIG. 5 is illustrated in FIG. 6. The illustrated maximum probability value determiner 240 includes: MUX1 242 as a first comparator for selecting a large value by comparing a transition probability value for (0,0) and a transition probability value for (0,1); If the selection value of the MUX1 242 is (0,0), the transition probability value for (1,0) is selected, and if the selection value of the MUX1 242 is (0,1), (1,1) A switch 244 as a switching unit for selecting a transition probability value for? And a MUX2 246 for selecting a large value by comparing the selection value of the MUX1 242 and the selection value of the switch 244.

The LLR calculator 260 obtains an LLR by applying one of Equations 16 to 19 according to the determination result of the maximum probability value determiner 240.

Detailed operations of the LLR calculator shown in FIG. 5 can be easily inferred from the method description of the present invention, and thus detailed description thereof will be omitted.

The present invention is more efficient than the LLR calculation method using a conventional complex formula by using the block processing to calculate the LLR by selecting one of a plurality of simplified LLR calculation formulas according to the maximum value of the transition probability values of each state. LLR calculation can be performed with

Claims (13)

(a) obtaining alpha, beta, and gamma values of two or more time intervals; (b) calculating transition probability values in the two or more time intervals using the alpha value, the beta value, and the gamma value; (c) performing a comparison operation on some of the transition probability values to determine a large value, selecting one value from the remaining ones of the transition probability values according to the determined value, and comparing the determined value with the selected value Obtaining a maximum value of the transition probability values in such a manner as to select a large value; And (d) determining an LLR by determining an operation to be applied according to the maximum value LLR calculation method comprising a. The method of claim 1, In the steps (a) and (b), the LLR calculation is performed on the values (0,0), (0,1), (1,0), and (1,1) for two time intervals. Way. The method of claim 2, wherein step (c) comprises: (c1) selecting a large value by comparing a transition probability value for (0,0) with a transition probability value for (0,1); (c2) If the transition probability value for (0,0) is greater in step (c1), select the transition probability value for (1,0), and if the transition probability value for (0,1) is larger, ( Selecting a transition probability value for 1,1); And (c3) selecting a larger value between the selection value of step (c1) and the selection value of step (c2) as a maximum value; LLR calculation method comprising a. The method of claim 3, wherein step (a) comprises: LLR calculation method characterized in that performed by applying the following equations.

The method of claim 3, wherein step (b) comprises: LLR calculation method characterized in that performed by applying the following equations.

The method of claim 3, wherein step (d) And performing a selected one of the following equations according to the maximum transition probability value obtained in step (c).

A forward state metric (FSM) calculation unit for obtaining an alpha value in at least two time intervals; A reverse state metric (BSM) calculation unit for obtaining a beta value in at least two time intervals; A branch metric (BM) calculation unit for obtaining a gamma value in two or more time intervals; A transition probability value calculating unit for obtaining a transition probability value in the two or more time intervals of each state from the alpha value, the beta value, and the gamma value; A maximum probability value determining unit for obtaining a maximum value of the transition probability values; And An LLR calculator configured to calculate an LLR by performing an operation specified according to the maximum probability value with respect to the transition probability values. LLR calculation device comprising a. The method of claim 7, wherein the maximum probability value determination unit, A first comparator for selecting a large value by comparing a transition probability value for (0,0) and a transition probability value for (0,1); If the selection value of the first comparator is (0,0), the transition probability value for (1,0) is selected, and if the selection value of the first comparator is (0,1), for (1,1) A switching unit for selecting a transition probability value; And A second comparator for selecting a large value by comparing the selection value of the first comparator with the selection value of the switching unit; LLR calculation device comprising a. The method of claim 7, wherein the branch metric calculation unit, Apparatus for calculating a gamma value in two or more time intervals by applying the following equation.

The method of claim 9, wherein the forward state metric calculation unit, LLR calculation device characterized in that to obtain the alpha value in two or more time intervals by applying the following equation.

The method of claim 9, wherein the reverse state metric calculation unit, LLR calculation device characterized in that to obtain a beta value in two or more time intervals by applying the following equation.

10. The method of claim 9, wherein the transition probability value calculation unit, LLR calculation device characterized in that to obtain a transition probability value by applying the following equations.

The method of claim 9, wherein the LLR calculator, LLR calculation apparatus characterized in that performed by applying the selected one of the following equations according to the maximum transition probability value determined by the maximum probability value determination unit.

Images