EP1374417A1

EP1374417A1 - Channel coding method

Info

Publication number: EP1374417A1
Application number: EP02714020A
Authority: EP
Inventors: Wen Xu
Original assignee: Siemens AG
Current assignee: Siemens AG
Priority date: 2001-04-05
Filing date: 2002-02-15
Publication date: 2004-01-02
Also published as: US7032152B2; DE10117034A1; US20040100995A1; CN1255953C; WO2002082662A1; CN1500309A

Abstract

The invention relates to a method for channel coding a parameter whose values are correlated with one another to different extents and according to which code words with better distance properties are at least partially associated with the more strongly correlated values of the parameters, and code words with weaker distance properties are at least partially associated with the more weakly correlated values of the parameter.

Description

description

Channel coding method

The invention relates to a method for channel coding of a parameter, the values of which are correlated with one another to different degrees.

The rapid technical development in the field of mobile communication has led to the development of so-called adaptive multirate (AMR) narrowband and wideband voice codecs in recent years, which are to be used in future mobile radio standards such as GSM / EDGE and UMTS. An AMR codec makes it possible to adaptively adjust the total data rate of the data to be transmitted, as well as the distribution of the data rate to the source and channel coding, depending on the channel status and network conditions (system load). The main objectives of such AMR voice codecs are to achieve fixed network quality of the voice under different channel conditions and to ensure optimal distribution of the channel capacity taking into account certain network parameters.

An AMR has several codec modes (modes) that indicate different rates of the coded speech bits. For a limited transmission bandwidth, an AMR should work in high-rate modes under good channel conditions and / or in highly utilized radio cells. It should be switched dynamically to the low-rate modes under poor channel conditions. This should result in the best speech quality taking into account the changing channel conditions.

In the GSM / EDGE system, 2 bits (mode bits), which form the so-called identifier, and up to 4 different modes of the AMR codec are reserved in each speech frame of 20 ms (both in the full rate and in the half rate channel)

(CODEC_MODE_l, CODEC_MODE_2, CODEC_MODE_3 and CODEC_MODE_). The 2 bits id (l), id (0) of the identifier are encoded using a block code. In this case, each Para ^¬ meter value are assigned a code word. The amount of the signaled by the identifier (maximum 4) modes is referred to as Ac ^¬ tive codec set of 8 possible in principle for Narrowband AMR codec modes. At least this active codec set is known to the base station and the mobile station. Since the 2 mode bits are transmitted together with the coded speech bits within a channel, they are also referred to as in-band data (see Table 1 and Table 2).

Table 1 Channel coding of the parameter "Identifier * in Adaptive multi rate speech Channel at fill rate (TCH / AFS) (excerpt from GSM 05.03, Section 3.9)

Table 2 Channel coding of the parameter "Identifier * in Adaptive multi rate speech Channel at half rate (TCH / AHS) (excerpt from GSM 05.03, Section 3.10)

On the receiving end, this identifier, also referred to below as a parameter, is first decoded, and in Furthermore, depending on this decoding result and thus the current codec mode, the speech bits are determined. If the parameter and thus the codec mode are incorrectly recognized, for example because of a very noisy received signal, the speech frames are decoded incorrectly and the speech information can be lost.

The initial codec mode during call setup or handover is defined as follows:

If all 4 codec modes are used, then the initial codec mode is the codec mode of the active codec set with the second lowest rate; otherwise the codec mode with the lowest rate. In any case, the codec mode changes from frame (k-1) to frame k only to one of its "neighboring" codec modes or remains in the same codec mode (temporal correlation of the values of the parameter). A jump to a non-neighboring mode, e.g. from CODEC_MODE_l to CODEC_MODE_3 is prohibited (see table 3). In other words: the parameter "identifier * has memory or is correlated from frame to frame (inter-frame correlationo the temporal correlation). The memory of the parameter "Identifier * leads to a limited change in the codec modes, or equivalent, the code words (CW _lf ..., CW) from frame (frame) (k-1) to frame k.

Frame (k-2) Frame (k-1) Frame k

Table 3 The present invention is therefore based on the object of specifying a method for channel coding of a temporally or locally correlated parameter which enables the parameter to be transmitted reliably and at low cost.

This object is achieved by the features of the independent claims to ^¬. Advantageous and expedient further developments result from the dependent claims.

The invention is therefore based on the idea of taking into account the correlation or the "memory" of the values of the parameter when coding the parameter, so that this can be used in the decoding. For this purpose, the values of the parameter are mapped onto code words in accordance with the correlation of the values of the parameter. In particular, a parameter is coded or its values are mapped onto code words in such a way that code words with better distance properties are assigned to “neighboring” or more strongly correlated parameter values. In the example above, these are the parameter values to which you can switch directly from the current parameter value. Codewords with poorer distance properties are assigned to the weakly correlated (“non-adjacent *) parameter values.

Within the scope of the invention, the parameter is also to be understood as any size that can have any values. A “frame *” can also be understood to mean a point in time or a location. Since a code word corresponds exactly to / is assigned to a parameter value, the following will often no longer differentiate between the terms “parameter value * and“ code word *.

The invention makes it possible to achieve an improved overall performance compared to known channel codes, which are optimal for parameters without thinking, without increasing the complexity (computation effort and memory requirement). The invention is described in more detail below on the basis of preferred exemplary embodiments, in which the channel coding is carried out using block codes, but this is in no way necessary to implement the invention; Rather, the invention is in regard to the present ^¬ message to many other types of encoding, in particular the channel coding, transferable.

A. Distance properties of channel codes (codes)

A channel code usually has a number of code words. Any part of the code words of a code (mother code) also form a subcode, with all code words of the subcode also being code words of the corresponding mother code. Distance (two code words) is often also understood to mean the Hamming distance, i.e. the number of bit positions at which the values of the two code words are not equal. Further distances such as the Euclidean distance of two code words can also be used in the present invention.

The set of distances of all possible pairs of code words of a code form a distribution {A (w), for w = d _mιn , d _mιn + 1, ... with a minimum Hamming distance d _mιn ), where A [w) is the An - Number or the (standardized) relative frequency of the pairs of code words with distance w indicates. For a normalized relative

Frequency holds ∑ (w) = l, and the mean distance is ^w = ^d mm ^d «= ∑ ^wA (w) ^w = ^d mm

As a rule, a code has a better distance characteristic than another code with the same code rate and the same dimension of code words if the number of code words in the code occupy larger minimum and / or medium minimum distances. In particular, the distance property can be assessed according to one of the following criteria: A code C with a minimum Hamming distance d _ιn and a relative frequency {A '(w)} has a better distance _property than a code C with a minimum Hamming distance d _min and a relative frequency {A (w)} _f if d mm - ^ O-now or if d _ιn = d _mιn and d ' _av = ∑wA'(w)> d _av = ∑wA (w). ^{w =} "now ^{w =} " πw

- A code C has a better distance property than one

Code C if Ö min 'Cπun or if d' _m2 _. = _Λ and

A '(w) = A (w) for w = d _mιnf d _mιn + 1, ..., w ₀ {w ₀ ≥ d _mιn ) and A' {w ₀ ) <A {w ₀ ).

Distance properties of a code such as d "u _n and d _av are fundamental in determining the ability of the error correction. Codes with better distance _properties , ie larger d _min and d _av , also have better correction _ability . So far, mostly good codes have been designed on the assumption that the data to be encoded has no memory. If there is memory in data or a parameter, such codes are usually no longer optimal.

As shown in Tables 1 and 2, the parameter indicating the codec mode is block encoded. For SID (Silence Descriptor) and RATSCCH (Robust AMR Traffic Synchronized Control Channel) frames there are code words with a length of 16 bits: CW _m = ic (15), ..., ic (0) with m = 1, 2, .. ., 4. For the speech frames there are code words with a length of 8 bits in the full rate channel:

CW _m = ic (7), ..., ic (0) in the full rate channel and code words with a length of 4 bits in the half rate channel:

CW _m = ic (3), ..., ic (0).

The simple case of a speech frame in the half-rate channel (Table 2) is discussed below. The Hamming distances between all 4 different code words are shown in Table 4.

CW _λ = 0000

Code word cw ₂ = 1001 cw ₃ = Olli

CW _ä = 1110 d (CW CIV ₂ ) = 2 d (CW _lf CW ₃ ) = 3

Hamming-d (CW _lf CN ₄ ) = 3

Distance d (CW _2r CW ₃ ) = 3 d {CW ₂ , CW ₄ ) = 3 d (CW ₃ , CW ₄ ) = 2

Table 4 Hamming distance of the code words of the block code for the speech frames in TCH / AHS.

B. Decoding the parameter without / with memory A simple approach to decoding the parameter is to calculate the correlation between the code words and the received mode bits, which are represented as the input vector of the channel software values. The mode (codec mode) is selected in which the correlation with the input vector is maximum (maximum likelihood decoding).

In the following example, a vector with 4 software values r = (r ₃ , r ₂ , r _lr r ₀ ,), each of which is quantized with 8 bits, is received for the coded mode bits. A value of r = +127 corresponds to a securely received "0 ^* and a value of r _λ = -127 corresponds to a securely received" 1 '.

Now the correlations corr _m (= correlation between r and CW _m ) are calculated for each mode m = 1, ..., 4, whereby the code word CW _m with "+ 1V" -1 * instead of "0 * /" l * (e.g. in Table 4).

This results in: corr - + r ₃ + r ₂ + r, + ^r o corr ₃ = + r ₃ - r ₂ - r ₎ ^{~ r} o

The mode of operation is illustrated below using a specific example. CIV ₂ = 1001 is sent. For the ideal channel, the receiver receives the vector r = (-127, +127, +127, -127). The result is: corr _λ = 0 corr ₂ = +508 = maximum corr ₃ = -254 corr _A = -254

The decoder correctly decides on the mode CODEC_MODE_2

(ie CW ₂ ).

Under a noisy channel it is also possible to correct certain errors. For example, if the vector is received for CW ₂

(r ₃ , r ₂ , r _lf r ₀ ,) = (-6, +2, +3, +5), whereby the bold “+ 5 *” bit has been rotated, the following applies: corr = +4 corr ₂ = +6 = maximum corr ₃ = -16 corr = +6 = maximum

In this case, a decision between C1V ₂ and CfV _{4 is} no longer possible. However, if the parameter memory is taken into account, e.g. if in the last frame {k- 1) CODEC_MODE_l (= CWi) or CODEC_MODE_2 (= CW ₂ ) is already decoded with a high level of security, then the decoder can correctly choose CW ₂ , since a change from CWi or CW ₂ to frame k (1) to the current frame k CWΛ is not allowed. In this case, the calculation of corr _{4 is} actually not necessary.

C. Distance properties for parameters without / with memory If all codec modes (i.e. all code words) are possible at time k during decoding (i.e. no memory available), the minimum Hamming distance results for code C with 4 code words in Table 4

Qmi n ^ and the relative frequency A (w) = 2/6 for w = 2

A (w) = 4/6 for w = 3

A (w) = 0 otherwise therefore d _αv = ∑wA (w) = 1616 ^w = ^d m

If the parameter at time (k-1) is in CODEC_MODE_l, then only CODEC_MODE_l and CODEC_MODE_2 at time k are possible (see Figure 1). This corresponds to a subcode with only two code words CWi and CW ₂ , where O _m _ι n ⁼ -

A (w) = 1 for w = 2 A (w) = 0 otherwise d _αv = ∑ wA (w) = 2

• "" " 'mm

When decoding, only the correlations corri and corr ₂ need to be taken into account.

In practice there are other typical situations: CW ₂ is the actual code word at time (k-1) and CW _{3 is} the actual code word at time k, but CW ₂ is decoded as CWi at time (k-1). In order to be able to decode the code word as CW ₃ at time k, uss CW ₃ as code word in Subcode be included, the k at the time to be taken into account when Decodie ^¬ tion. A change from CWi to CW ₃ should also be allowed to avoid further decoding errors. However, the probability of such a change is very small compared to a normal change such as CW ₂ to CW ₃ . Furthermore, it is conceivable that even a change from C i to CWΛ should be possible because of several decoding errors in succession. However, the probability of such a change is so low that it can be neglected. For this reason, the distance d (CW _x , CW ₄ ), which is crucial in decoding to distinguish CW _X and CW ₄ , is not taken into account. To take this effect into account, in our example we define a special subcode that has all the code words of the mother code {CWi, CW ₂ , CW ₃ , CW ₄ }, but the distance d (CW _lf CW) is neglected when considering the distance _properties . The distance properties of this subcode, which are generally not identical to those of the mother code, can also be used when decoding the parameter with memory.

In general, for parameters with memory not only the distance property of the mother code for parameters without memory are decisive, but also the distance properties of all such subcodes that correlate with the groups of the strongly

Parameter values are assigned. Table 5 shows the results of d _min and d _{av of} all subcodes for possible known modes at time (k-1).

It should be noted that the different subcodes (or the corresponding groups of parameter values) can overlap, ie the same code words exist in different subcodes. As Table 5 shows, CW ₂ belongs to the subcodes {CWi, CW ₂ } (for CODEC_MODE_l at time (k- 1)), {CW _X , CW ₂ , CWi) (for CODEC_MODE_2), etc.

D. Find the best codes for parameters with memory For parameters with memory, the ability of the error correction ^¬ is mainly determined by distance properties of the individual sub-code. Therefore, should the code words for the Pa ^¬ rameterwerte selected to be optimum distance properties that arise (for example largest d _min and d _av) of all Subco ^¬ of which are assigned to the groups of the correlated parameter values.

Table 5 Distance properties of the subcodes of a specified block code for decoding the "identifier * for speech frames in TCH / AHS.

Table 6 Distance properties of the subcodes of an optimized block code (with CW \ = CW _lf CW ' ₂ = CW _4f CW' ₃ = CW _3t CW '* = CW ₂ ) for decoding the "identifier" for speech frames in TCH / AHS.

Is precise information about the probability of the code word change CWi (λ-l) → CW _m (k) (= change of the parameter value) from frame (k-1) to frame k, namely P (CW _m (k) \ CW _λ ( k- 1)), available (1, m, n = 1, 2, ...), this information can be taken into account in the code design.

If P (CW _n (k) | CW _X (k-1)) ≥ P (CW _m (k) \ CW (kl)), then the code words CW _lf CW _m and CW _{n should be} generated such that for the Hamming distance d (CW _nι CW _λ ) ≥ d (CW _mt CW ₂ ) applies. This is a general mathematical formulation of the present method if all code words CWi, CW _m and CW _{n are determined} in this way. For mutually exclusive code words CW ₂ and CW _m , P (CW _m (k) \ CW (k-1)) = 0. However, d (CW _m , CW)> should normally be> 0, otherwise CW _m and CWj become a single code word.

The parameter is, for example, in the frame (k-1) CW ₂ (see table 3). If P (CW ₃ (k) | CW ₂ (k-1)) ≥ P (CWXk) \ CW ₂ (kl)) ≥ P (CW (k) | CWι (λ-l)) = 0 (since CW _X and CW ₄ are mutually exclusive), then CWi, CW ₂ , CW ₃ and CW _{4 should be} generated such that d (CW ₃ , CW ₂ ) ≥ d (CW _x , CW ₂ ) ≥ d (CW _λ , CW ₄ ) applies.

If the best codes are then selected in accordance with the criterion explained above, it is noticeable that there are generally several codes which have identical distance properties. In the case of block codes, such codes can be generated, for example, by adding any vector to all code words or / and by simultaneously exchanging any bit positions for all code words. Table 6 shows the distance properties of a code optimized according to parameter memory with code words CW'i = CW _X , CW ' ₂ = CW _{4 r} CW ^f ₃ = CW ₃ , CW' i = CW ₂ . This code has a higher ability to correct errors than the block code in table 2. If the mode CODEC_MODE_l or C0DEC_M0DE_4 is in frame (k-1), the subcode of the optimized block code has a minimum Hamming distance of 3, with which even with hard ones Decision an error can be corrected. In contrast, the subcode of the specified block code with a minimum Hamming distance of 2 cannot correct an error if the decision is hard.

E. Assignment of the code words according to parameter memory The analysis above shows that due to the possibility of several successive decoding errors - although the probability is very low - not only the individual subcodes, but also the mother code should have the best possible distance properties in order to achieve the best possible performance. This is particularly important if there is no strong parameter correlation.

In the present application, a method is proposed in which the code which is optimal for the memoryless parameter, in particular its minimum Hamming distance (d _min ), is first used, and then the code words are specifically assigned to different parameter values. Specifically, as shown in the previous example, the code words are assigned so that the subcodes have the best possible distance properties for individual groups of the correlated parameter values. In this way, the optimized code achieves good performance for parameters without memory as well as for parameters with memory.

If more precise information such as P (CW _m (k) \ CW _λ (k-1)) is available, this information can also be taken into account when assigning the code words. If P (CW _n (k) \ CW ₂ (k-1)) ≥ P (CW _m (k) \ CW _λ (kl)), the code words CWj ,, CW _m and CW _n , if possible, according to the criterion d (CW _n , CW _λ ) ≥ d (CW _mi CW) assigned. Since in this case all code words are already defined, we can first sort P (CW _m (k) | CW _X (k-1)) and d (CW _ml CW _λ ) (1 ≠ m) according to their size, and finally assign the appropriate code words to the parameter values.

As shown in Table 4, the minimum Hamming distance is achieved with C i and CW ₂ (or CW ₃ and CW ₄ ). Therefore, CW _λ and CW ₂ (or CW ₃ and CW ₄ ) should be assigned to the mutually exclusive parameter values C0DEC_M0DE_1 and C0DEC_M0DE_4. If the parameter values C0DEC_M0DE_1 and C0DEC_M0DE_4 are assigned, for example, CWi and CW ₂ , there are also parameter values C0DEC_M0DE_2 and C0DEC_M0DE_3, to which CW ₃ and CW ₄ can be assigned.

In this way, one obtains (by chance) the same results in Table 6, the subcodes {CW'i, CW ' ₂ }, [CW' _3l CW ^f ₄ ) _having the better distance _properties with d _mιn = 3 than the subcodes {CW _lf CW ₂ ), [CW ₃ , CW ₄ ) with d _mιn = 2. The performance for parameters without memory remains unchanged.

This procedure can be used to optimize the codes for language frames in TCH / AFS (see tables 7 and 8) and for SID and RATSCCH frames (see tables 9 and 10).

Table 7 Distance properties of the subcodes of the block code specified in GSM 05.03 for decoding the "identifier * for speech frames in TCH / AFS.

Table 8 Distance properties of the subcodes of an optimized block code (with CW ' _X = CW _lr CW' ₂ = CW ₄ , CW ' ₃ = CW ₃ , CW' ₄ = CW ₂ ) for decoding the "Identifier * for speech frames in TCH / AFS.

Table 9 Distance properties of the subcodes of the block code specified in GSM 05.03 for decoding the "Identifier * for SID and RATSCCH frames.

Table 10 Distance properties of the subcodes of an optimized block code (with C 'i = CW _X , CW' ₂ = CW ₂ , CW ' ₃ = CW ₄ , CW'i CW ₃ ) for decoding the "Identifier * for SID and RATSCCH frames.

reference

[1] GSM 05.03, "Digital cellular telecommunications System (Phase 2+); Channel coding," Release 1999, 2000. [2] GSM 05.09, "Digital cellular telecommunications System (Phase 2+); Link Adaptation," Release 1999, 2000th

Claims

claims

1. Method for channel coding of a parameter, in which the values of the parameter are correlated to different extents or times, in which values of the parameter are mapped to code words,

in which code words have different distances from one another, and in which values of the parameter are mapped to code words in such a way that values of the parameter which are relatively strongly correlated with one another are mapped to code words which have relatively large distances from one another.

2. method for channel coding of a parameter,

- in which a first value of the parameter and a second value of the parameter form a pair of values, the second value of the parameter depending at least partially on the first value of the parameter, - in which values of the parameter are mapped to code words,

- With the code words have different distances from each other, and

- In which values of the parameter are mapped onto code words in such a way that a pair of values whose second value depends more strongly on the first value is mapped onto code words that are at greater distances from one another than a pair of values whose second value depends less on the first value ,

3. Method for channel coding of a parameter in which the values of the parameter are correlated with one another in time or location, and in the case of code words with better distance properties at least partially the more correlated values of the parameter, and code words with poorer distance properties at least partially the weaker correlated values of the parameter.

4. The method as claimed in one of the preceding claims, in which mutually exclusive values of the parameter are assigned code words with relatively poor distance properties.

5. The method according to any one of the preceding claims, in which a channel coding is used which is optimal for parameters without memory, in which in particular a channel code is used which is optimal for parameters without memory.

6. The method according to any one of the preceding claims,.

- In which a channel coding is used in which the minimum Hamming distance is maximum, in which in particular a channel code is used whose minimum Hamming distance is maximum.

7. The method according to any one of the preceding claims, - in which the channel coding is carried out using a block code.

8. The method according to any one of the preceding claims, wherein the parameter is used to signal the mode of an adaptive multi-rate codec.

9. A method for decoding a parameter which has been channel-coded according to one of the preceding claims, in which the correlation of the parameter is taken into account.