CN101262233A

CN101262233A - Viterbi coding method

Info

Publication number: CN101262233A
Application number: CNA2008103011816A
Authority: CN
Inventors: 杨璐
Original assignee: Sichuan Hongwei Technology Co Ltd
Current assignee: Sichuan Hongwei Technology Co Ltd
Priority date: 2008-04-17
Filing date: 2008-04-17
Publication date: 2008-09-10
Anticipated expiration: 2028-04-17
Also published as: CN101262233B

Abstract

The invention relates to a decoding technology, particularly to a Viterbi decoding method used in the digital audio and video broadcasting system adopting an OFDM (orthogonal frequency division multiplexing) communication method. The Viterbi decoding method of the invention can reduce the complexity by reducing a survivor path with low likelihood. The Viterbi decoding method comprises the steps of: a. the quantity of original states are set; b. status measurement is calculated according to the value of CSI, and the value of CSI is used as the sum of corresponding Euclidean distance of the present-moment states of the weighting coefficient; c. the quantity of hold states is confirmed according to the values of CSI, and bigger value of CSI leads to less quantity of hold states; d. branching measurement is calculated; e. the corresponding branching measurements of the hold states are calculated, and path selection is conducted according to the branching measurements; f. that whether the decoding process is finished or not is judged, wherein, the coding is output when finished, and the step b is reduced when the decoding process is not finished. The Viterbi decoding method of the invention can reduce the complexity of the Viterbi decoding process and the possibility of overflow of a stored buffer.

Description

Viterbi coding method

Technical field

The present invention relates to decoding technique, particularly adopt the digital audio of OFDM (OFDM) communication, the Viterbi decoding in the video broadcast system.

Background technology

Traditional AM, FM analog broadcasting are disturbed irresistance for multipath, and the audio program quality that causes receiving descends; Digital audio broadcasting (DAB) then possesses superior high speed action receiving ability and signal quality, and the single frequency network technology of employing can allow the listener use same frequency to listen between the unlike signal block, and can receive the multimedia messages except that audio frequency.It is the third generation broadcasting after Frequency And Amplitude Modulation broadcasting.

The DAB receiver is divided into high intermediate frequency front end, base-band digital processing, audio coding and control four parts.Base-band digital is handled the DAB data is carried out disposed of in its entirety.The DAB data channel is divided into synchronizing channel, snap information passage (FIC) and main business passage (MSC), and MSC holds multichannel voice frequency and data service simultaneously.The demultiplexing circuit separates FIC with MSC, and according to subscriber control information chooser passage in MSC.Separate time domain and interweave and only be used for MSC, error-correcting decoding then carries out respectively FIC and MSC, has adopted Viterbi (Viterbi) Soft decision decoding.The control section of receiver is sent in FIC data after the decoding and MSC data service, and decoding back MSC voice data is sent into the mpeg audio decoder after the speed adjustment.Pcm stream after the decoding is exported behind D/A.Compared with former analog broadcasting, the signal strength signal intensity the during transmission of DAB signal is very little, particularly mobile environment.Therefore require received signal that the DAB receiver can maximum possible and the error code in the transmission is corrected, this is the key factor that influences DAB receiver quality.

Viterbi interpretation method (VA) is a kind of probabilistic decoding method of the best, is from 2 ^M-1Plant the effective calculation method of the conditional probability of the most probable bit sequence that upgrades optimum state and transmission in (m is the encoder limited length) possibility state,, must calculate 2 each for calculating this probability ^M-1Kind of state is stored in the path memory as a bit position according to each final decisions of making of these calculating.Cataloged procedure be equivalent in fact on the grid chart of sign indicating number, need look for one with the corresponding path of source sequence process, decode procedure then is to seek a process with the nearest path of receiving sequence on the grid chart of sign indicating number.In Viterbi decoding, two kinds be can use and path metric (Euclidean distance) and branch metric measured.Branch metric is the pairing Euclidean distance of current time state (Euclidean distance) sum.

As illustrated in fig. 1 and 2, in VA, Branch Computed is measured behind the setting initial condition number; All might turn to its follow-up two kinds of in may states one to every kind of state, each is corresponding to the bit of up-to-date input in the convolutional encoder shift register, and this bit is 0 corresponding a kind of state, or is 1 corresponding a kind of state; The Euclidean distance (path metric) of two paths on each state node of grid chart is compared, and preserve in the path (path that probability event is higher) that will have less Euclidean distance, and another paths is abandoned, and stores corresponding branch metric value simultaneously.Decoder judges that by Jia-ratio-choosing (ACS) operation any is possible state.Each path metric that addition refers to previous level is added in two branches that allow to shift, relatively refer on given level the branch metric of the branch that enters a certain state is compared, selection refers to selects higher one of probability event among both, abandons lower one.Because each state node draws two branch roads, therefore the extension in the 2nd grade of later paths at different levels all doubles, but through relatively and after selecting, has abandoned half, and the result retains the total number of paths of getting off and keeps constant (the survivor path number equals the initial condition number).Storage and processing for survivor path have two kinds of diverse ways, a kind of crying " register exchange exchange " method, a kind of crying " recall in the path " method.

As shown in Figure 1, " register exchange " method need all be stored a survivor path sequence to each state.Every one step of processing all will exchange survivor path by comparative result, and adds a new decision value and will organize a decision value output that deposits in earlier in a side of path sequence.When high-speed coding, these action needs are finished by hardware is parallel.In low speed decoding be, these operations can be finished in serial, but time-delay is bigger.

As shown in Figure 2, " recall in the path " method is applicable to the decoding of higher rate, because it is faster than the serial realization speed of register exchange method.This method does not need to store actual discriminative information sequence, and generation during with storage opposite order return rope, thereby find out the maximum-likelihood decoding sequence, can repeatedly decipher at high speed a plurality of branch roads.

The complexity of Viterbi is mainly reflected in the variable storage problem in amount of calculation and the decode procedure.

Summary of the invention

Technical problem to be solved by this invention is to provide a kind of by taking to reduce the Viterbi interpretation method that the low survivor path of likelihood score reduces complexity.

The present invention is for solving the problems of the technologies described above, and the technical scheme that is adopted is: viterbi coding method may further comprise the steps:

A, setting initial condition number;

B, measure according to CSI value computing mode: with the CSI value as the pairing Euclidean distance sum of the current time state of weight coefficient;

C, determine the reserved state number according to the CSI value, the status number of the big more reservation of CSI value is few more;

D, Branch Computed are measured;

The branch metric of e, calculating reserved state correspondence carries out Path selection according to branch metric again;

Whether f, decoding are finished, and in this way, decoding output if not, is returned step b.

Further, described step b is further comprising the steps of before computing mode is measured: judge the surplus mode of deleting according to deleting complement vector; Step b also measures according to deleting surplus mode computing mode: with the CSI value as deleting of weight coefficient the pairing Euclidean distance sum of state after surplus, the use of deleting complement vector during compute euclidian distances makes and separates that to delete surplus convolution code easier; The scope of the reserved state number of determining according to the CSI value is 1 to the initial condition number.

The invention has the beneficial effects as follows, can reduce the possibility of overflowing of the buffer of complexity in the Viterbi decode procedure and storage.The channel condition information that the CSI value is comprised can be used for adding than selecting process of Viterbi decoding, is used to increase amount of information, improves the confidence level of state measurement.

Figure of description

Fig. 1 register exchange method

Fig. 2 path backtracking method

Fig. 3 Viterbi interpretation method of the present invention

Fig. 4 coder state transition diagram

The 6th grade of grid chart of Fig. 5

Embodiment

CSI (channel state information, channel condition information) is usually used in the Viterbi decoding adaptive control of convolution decoder in the ofdm system.Specifically digital audio broadcast system (Digital Audio Broadcast, DAB) the middle use.

In order to reduce the complexity of Viterbi decoding, take to abandon the way in the low path of likelihood score.In the present invention, in order substituting all states to be judged, thereby to have been taked this than the maximum-likelihood method of substitution.When the status transmission of channel was fine, survivor path can seldom when status transmission is bad, can be adjusted the survivor path number, such as the big threshold value of accent, otherwise was easy to cause the path to be dropped, and formed error code.In order to realize the adaptive decoding of Viterbi decoding, utilize the CSI value to calculate the reserved state number, the big more explanation status transmission of CSI value is good more, and the status number of reservation is few more, promptly only keeps 2 ^M-1The higher partial status of " likelihood " degree in the individual state, so complexity reduces and the possibility of overflowing of the buffer of storage reduces greatly.Also introduce the weight coefficient of CSI value as the pairing Euclidean distance of current time state when computing mode is measured, the present invention is very approaching with existing Viterbi interpretation method error correcting capability, but amount of calculation obviously reduces.

The register exchange method need read the content of each state and import when upgrading survivor path, therefore brings a large amount of power consumptions.So preferential " (trace back) recalled in the path " method that adopts.In addition, it all is to replenish bit that existing deconvolution is deleted surplus method, and the method for invention can be when calculating path be measured, and decides the Euclidean distance of current path according to deleting complement vector.Just calculate the distance of current residual bit according to the log-likelihood ratio information of front end, rather than the distance of all bit.Deleting in the surplus viterbi of the being put into decode procedure, and need not reduce complexity separately to deleting surplus the processing to Puncturedconvolutional coding (deleting surplus convolution code) decoding.

" recall in the path " Viterbi interpretation method after the present invention improves, step as shown in Figure 3, may further comprise the steps:

A, setting initial condition number;

B, described step b also measure according to deleting the complement vector computing mode: with the CSI value as deleting of weight coefficient the pairing Euclidean distance sum of state after surplus;

C, calculate the reserved state number according to the CSI value, the status number of the big more reservation of CSI value is few more;

D, Branch Computed are measured;

E, carry out Path selection according to reserved state number and branch metric;

Whether f, decoding are finished, and in this way, recall in the path, and decoding output if not, is returned step b.

Embodiment

For the employed encoder limited length of DAB is 3 encoder, and there are 4 kinds of states in the memory of this encoder, just by its 2 bit register show 2 ²The state of kind.Corresponding Viterbi decoder exactly will be after receiving transfer sequence, determine in these states each state transitions to 4 kind of encoder may next group of state probability, as shown in Figure 4: (current state is S0, may turn to S0 or S2 in NextState all might to turn to its follow-up two kinds of in may states one to every kind of state; Current state is S1, may turn to S0 or S2 in NextState; Current state is S2, may turn to S1 or S3 in NextState; Current state is S3, may turn to S1 or S3 in NextState), each shifts the bit corresponding to up-to-date input in the convolutional encoder shift register, this bit is 0 corresponding a kind of state, or be 1 corresponding a kind of state, as shown in Figure 4, it is 0 that state S0 turns to the corresponding bit of S0 at NextState; To be S0 turn to the corresponding bit of S2 at NextState to current state is 1.

The information sequence of importing this encoder is (101110100), and the sign indicating number sequence of being exported by encoder is (11,01,00,10,01,10,00,01,11), deletes complement vector for (1110), delete through convolution and be output as (11 after surplus, 0,00,1,01,1,00,0,11) behind the DMC channel, the sequence of sending into decoder is the sequence that four level quantize _R(3 2, 0, 00, 2, 0 1, 3,0 1, 0, 23), the band underscore character is illustrated in the character that makes a mistake in the transmission, and the CSI value is 1 expression inerrancy, and the more little expression error rate of CSI value is big more, and corresponding C SI value is (1,0.85,0.5,1,1,0.8,0.95,0.6,1,1,0.8,1,0.85,1).Sequence after four level quantize if do not make a mistake should be (33,0,00,3,03,3,00,0,33), and is as shown in table 1:

Information sequence	1	0	1	1	1	0
Information sequence	1	0	1	1	1	0	Encoder output	11	01	00	10	01	10
Convolution delete surplus after	11	0	00	1	01	1	Encoder output	11	01	00	10	01	10
Convolution delete surplus after	11	0	00	1	01	1	Four level quantized sequences R	3 2	0	30	2	0 1	3
CSI	1，0.85	1	0.5，1	0.8	0.95，0.6	1	Four level quantized sequences R	3 2	0	30	2	0 1	3
CSI	1，0.85	1	0.5，1	0.8	0.95，0.6	1	Standard four level quantize	33	0	00	3	03	3

Table 1

In the OFDM of DAB communication modes, by the communication bandwidth of design subcarrier, can regard each subcarrier as flat channel, so, utilize pilot channel to estimate the channel distortion of each subcarrier easily.Equally, also the channel distortion that can utilize this estimation to obtain removes to improve traditional Viterbi algorithm, and this also is a distinctive Viterbi decoding algorithm in the ofdm communication system.Therefore for the estimation of channel conditions, calculate by the CSI value, the channel condition information that CSI comprised simultaneously can be used for adding than selecting process of Viterbi decoding.Variance and signal to noise ratio according to channel transfer function determine the status number that keeps then the CSI value to be brought in the Viterbi decode procedure together.Utilize the CSI value to realize self adaptation, set threshold value here, when CSI satisfies certain condition, keep corresponding status number.Being provided with of this routine threshold value and reserved state number is as shown in table 2 below:

The CSI value	＞0.9	?0.8～0.9	?0.6～0.8	＜0.6
The CSI value	＞0.9	?0.8～0.9	?0.6～0.8	＜0.6	The reserved state number	1	?2	?3	?4

Table 2

The corresponding Euclidean distance of state transitions institute was as shown in table 3 below when computing mode was measured:

Table 3

Utilize the flow process of self adaptation Viterbi decoding of channel condition information as follows:

Step 1:

1-1 sets initial condition several 2 ^K-1(K=3); Promptly 2 ²=4;

1-2 judges the surplus mode of deleting; The complement vector of deleting of current bit correspondence is 11, does not just delete surplus;

1-3 measures according to CSI value computing mode, obtains current C SI (1)=1 according to table 1, CSI (2)=0.85, and R (1)=32, a last moment state S0 may turn to (S0 or S2) in its follow-up two kinds of possibility states, as shown in Figure 4.Status quantitavie for CSI value (1,0.85) as deleting of weight coefficient the pairing Euclidean distance sum of state (nothing is deleted surplus) after surplus (according to the

sequence

3,2 of the four level quantification that receives, can by table 3, when the bit of state transitions correspondence was 0, corresponding Euclidean distance was 0,5; When the bit of state transitions correspondence was 1, corresponding Euclidean distance was 10,0).(c) the expression laststate is the tolerance of the state a of b at moment c to status quantitavie S for a, b, two states in the first time during state transitions is measured and is respectively: S (0,0,1)=and 1*0+0.85*5=4.25, promptly laststate is the state measurement of the state S0 of S0 in the moment 1; S (2,0,1)=1*10+0.85*8=16.8, promptly laststate is the state measurement of the state S2 of S0 in the moment 1;

1-4 calculates reserved state number: CSI (1)=1, CSI (2)=0.85, and CSI (2) is less, therefore tables look-up according to this value and 2 determines reserved states, keeps two states (be because this is the starting point of grid chart, so keep current all two states here);

The 1-5 Branch Computed is measured: be starting point here, and branch metric BM (0,0,1)=S (0,0,1)=4.25, BM (2,0,1)=S (2,0,1)=16.8; BM (a, b, c) branch metric of expression moment c from state b to state a;

1-6 keeps the state that satisfies condition and arrives respective paths, P (0,0,1)=[S0], P (2,0,1)=[S2]; P (a, b, c) survivor path of the path correspondence of expression moment c from state b to state a;

Does 1-7 feed back?

, do not move to step 2;

Step 2:

2-2 judges the surplus mode of deleting; The complement vector of deleting of current bit correspondence is 10, has just only kept first bit.

2-3 measures according to CSI value computing mode, CSI (3)=1, and R (2)=0, the S0 state is S (0,0,2)=1*10=10 at the status quantitavie of current time, S (2,0,2)=1*0=0, S (1,2,2)=1*0=0, S (3,2,2)=1*10=10;

2-4 calculates the reserved state number; CSI (3)=1, the reserved state interval is 1, therefore keeps a state and gets final product.

The 2-5 Branch Computed is measured, branch metric BM (0,0,2)=4.25+1*10=14.25, BM (2,0,2)=4.25+1*0=4.25, BM (1,2,2)=16.8+1*10=26.8, BM (3,2,2)=16.8+1*0=16.8;

2-6 keeps the state that satisfies condition P (1,2,2)=[S2, S1] to respective paths;

Does 2-7 feed back?

, do not move to step 3:

Step 3: do not delete the bit number, obtain status quantitavie S (0,1,3)=5, S (2,1,3)=10, branch metric BM (0,1,3)=31.8, BM (2,1,3)=36.8 according to the CSI value; The reserved state interval is 4, keeps at present all and relates to state P (0,1,3)=[S2, S1, S0], P (2,1,3)=[S2, S1, S2] are arranged to respective paths; Decoding is not finished, and moves to step 4;

Step 4: the reserved state interval is 2, keeps 2 states that satisfy reserve to respective paths, P (1,2,4)=[S2, S1, S2, S1] is arranged, P (3,2,4)=[S2, S1, S2, S3]; Decoding is not finished, and moves to step 5;

Step 5: the reserved state interval is 3, keeps 3 states that satisfy reserve to respective paths, P (1,3,5)=[S2, S1, S2, S3, S1] is arranged, P (2,1,5)=[S2, S1, S2, S1, S2], P (3,3,5)=[S2, S1, S2, S3, S3]; Decoding is not finished, and moves to step 6;

Step 6: the reserved state interval is 1, branch metric BM (0,1,6)=58, and BM (2,1,6)=48, BM (1,2,6)=55.1, BM (3,2,6)=65.1, BM (1,3,6)=65.7, BM (3,3,6)=55.7 keeps 1 state that satisfies reserve (branch metric maximum) to respective paths, be P (1,3,6)=[S2, S1, S2, S3, S3, S1]; Decoding is finished, and according to the grid chart of the 6th grade of state transition diagram shown in Figure 4 as shown in Figure 5, decoding is output as 11,01,00,10,01,10, and the table of comparisons 1 is consistent with the encoder output sequence.

Claims

1. viterbi coding method is characterized in that, may further comprise the steps:

A, setting initial condition number;

B, with the CSI value as the pairing Euclidean distance sum of the current time state of weight coefficient as status quantitavie;

D, Branch Computed are measured;

2. viterbi coding method according to claim 1 is characterized in that described step b is further comprising the steps of before computing mode is measured: judge the surplus mode of deleting according to deleting complement vector.

3. as viterbi coding method as described in the claim 2, it is characterized in that described step b also measures according to deleting surplus mode computing mode: with the CSI value as deleting of weight coefficient the pairing Euclidean distance sum of state after surplus as status quantitavie.

4. viterbi coding method according to claim 1 is characterized in that, the scope of the reserved state number of determining according to the CSI value is 1 to the initial condition number.