DESCRIPTION

Apparatus and method for especially Nordstrom-Robinson code decoder

Technical Field

The invention regards to an apparatus for especially Nordstrom-Robinson code decoder according to pre- characterizing part of claim 1, and to a method for Nordstrom-Robinson code decoder.

In modern digital communication systems, especially wireless mobile systems like terrestrial broadcasting TV system (TV: television) , high attention is paid to multi-path echoes, which are caused by reflection, refraction and so on. In single-carrier systems, an adaptive decision feedback equalizer (DFE) is often used to cancel a inter-symbol interference (ISI) caused by the echoes. But in some severe environments, where strong echoes exist and SNR (signal to noise ratio) is very low, adaptive DFE may not work well . A structure, combining the equalizer with a channel code can be used to improve the quality of symbols decision, which will be fed back to a FBF part (FBF: feedback filter) of the equalizer. For example, in ATSC system (ATSC: Advanced

Television Systems Committee) Viterbi code is used. The drawback of the convolution code is that it may cause error spreading .

In China Terrestrial Digital Television Standard, there is one recommended mode in which a block code called Nordstrom- Robinson code (NR code) is applied with QPSK modulation (QPSK: quadrature phase shift keying) . NR code, found by Nordstrom and Robinson, is one of the most remarkable codes known. It is the unique binary code consisting of 256

codewords of length 16, such that any two of them differ in at least six positions. This is the largest number of codewords possible for this length and minimum distance.

China Terrestrial Digital Television Standard recommends two carrier modes, single-carrier mode and multi -carrier mode. In single-carrier mode, NR code is combined with the equalizer to improve the quality of symbol decision, as discussed above. In multi-carrier mode, it performs as an inner code.

Background Art

Fig. 5 shows a general structure of a prior art NR decoder 6 combined with an equalizer and a FEC decoder 7 (FEC: forward error correction) . An input signal or data, especially a received vector is inputted into a feed forward filter (FFF) 1. Data outputted out of the feed forward filter 1 are inputted into an input of an adder 2. Further, data outputted out of a feedback filter (FBF) 5 are inputted into a further input of the adder . Data added in adder 2 are outputted into a NR decoder 6 and into an error generator 3. The NR decoder 6 takes the soft symbols from adaptive DFE, and decodes the received vector. Then, the NR decoder 6 outputs the results to the feedback filter 5, to the error generator 3 and to the forward error correction decoder 7 (FEC decoder) . Further, a tap coefficient updater 4 is provided for updating tap coefficients for the feed forward filter 1 and the feedback filter 5.

As the decoding procedure of NR code is block based, there will be a delay in the decoder. That means the decoder should collect 8 symbols, which contain 16 bits, before decoding every time. Assuming NR decoder 6 takes k symbols time for decoding process, and hence, the feedback filter 5

part of the equalizer cannot compensate the k + 8 symbols' interference to the received symbol of the decoder. So the NR decoder 6 should comprise a unit to cancel the interference .

The generic decoding method of NR coded information is maximum likelihood decoding. The most direct approach to this method is to compute the Euclidean distance from the received vector for each codeword and then to select the codeword closest to the received one in Euclidean distance. And this is the best way to decode a binary block coded vector, but its implementation scale is very large and not sustainable sometimes.

The input of FEC decoder can be hard decision results or soft information. Hard decision results are just the Maximum Likelihood decoded symbols described above. However, the performance of the FEC decoder will be greatly degraded if the inputs are hard decision results. Another way is to calculate the soft information every symbol in NR decoder and output it into FEC decoder.

In Chinese patent "A time-domain equalizer combined with NR code", CN patent application number 200510026192.4, a log- MAP algorithm is proposed to generate the soft information for FEC decoder. But the drawback of this algorithm is that the complexity is very high for implementation.

At a first aspect Maximum likelihood decoding is used. Generally, in a NR decoder, the most direct approach of maximum likelihood decoding is to compute the Euclidean distance from the received vector for each codeword and then to select the codeword set to the received one in Euclidean space. The Euclidean distance of vector x and y is calculated as

Euclidean distance = |x - y| ²

The drawback of the solution described above is its high computational complexity. There are 256 codewords in the NR table, so it needs 512 real-multiplications for one received vector decoding.

At second there exists a Log-MAP algorithm. For AWGN channel (AGWN: additive white Gaussian noise) , the received symbol is y = as + n, with here a as impulse response of the channel, n as the additive white Gaussian noise, s as a transmitted symbol, and y as the output signal. Further, s, y are complex base band signals, i.e. y = yi+jy _q , s =s _s +js _q .

After equalizer, whatever time domain equalizer or frequency domain equalizer will compensate the channel impact to the received symbol, thus the output symbol will be y = (as+ή)la = s+nla with n as Gaussian variable with zero mean and variance σ ² . The soft metrics for each bit of M-QAM symbols (QAM: Quadrature Amplitude Modulation) is shown as follows. Where

Piylx _k =ϊ ⁾ _{f (} £ _{= U3λ} 5,6,7,8), here llr(k) means the log-likelihood ratio for k-th bit in constellation symbol, x* represents the k-th bit in transmitted symbol s

Since n is Gaussian noise , thus ,

The drawback of log-MAP algorithm is also the high complexity for implementation.

Technical Problem

It is an object of the invention to provide an apparatus and method having complexity reduced code decoder, especially Nordstrom-Robinson code decoder.

Technical Solution

This object is solved by an apparatus for especially

Nordstrom-Robinson code decoder having features according to claim 1, and by a method for Nordstrom-Robinson code decoder having features according to claim 11. Preferred aspects and embodiments are subject-matter of dependent claims.

Especially, there is provided an apparatus having a block code Decoder for decoding code coded information, a distance calculator adapted to receive inter-block interference cancelled symbol, data or vector, respectively, and intra- block interference compensated code vector from a interference canceller, to calculate the Euclidean distance between the two vectors, a decoder adapted to receive the Euclidean distances calculated by distance calculator and to choose a minimum distance and its corresponding code codeword and to output a decision code vector, and a soft information calculator adapted to receive the Euclidean distances from distance calculator and to calculate a log-

likelihood ratio.

Advantageous Effects

Especially, there is provided an apparatus, wherein the distance calculator is adapted to use approximate operation to calculate the Euclidean distances between the received vector and every compensated code vector. This reduces the hardware complexity. Further, such kind of approximate functions can be absolute function or curve approximation, etc .

Preferably, the decoder is adapted to collect all Euclidean distances of a set of Euclidean distances, and to find out the minimum one and the corresponding codeword, and to map the decoded codeword to the code vector. Especially, the decoder is adapted to collect 256 Euclidean distances, and to map the decoded codeword having 16 bits to the code vector .

The soft information calculator can be adapted to get all Euclidean distances of a set of Euclidean distances from Euclidean distance calculator for one code codeword and to calculate the log-likelihood ratio using max-log-MAP algorithm. Especially, the soft information calculator is adapted to get 256 Euclidean distances from Euclidean distance calculator for one Nordstrom-Robinson code codeword. Using max-log-MAP algorithm is to reduce the computation complexity.

The interference canceller can comprise an inter-block interference canceller adapted to receive code vector and tap coefficients of feedback filter from equalizer and previous decoded code vector from decoder and to cancel the inter-block interference of the received code vector, and an

intra-block interference compensator adapted to get codewords from code look-up-table and tap coefficients of the feedback filter from equalizer, and to generate the compensated code vectors. Especially, such inter-block interference canceller cancels the inter-block interference of the received code vector, and the intra-block interference compensator is adapted to get codewords from code PN look-up-table and tap coefficients of the feedback filter from equalizer, to generate the compensated code vectors.

Especially, the inter-block interference canceller is adapted to collect tap coefficients of the feedback filter part from equalizer and previous decoded code vector and to operate convolution between them to generate inter-block interference, and to subtract the inter-block interference from the received code vector.

The intra-block interference compensator can be adapted to get codewords from code look-up-table and to operate convolution between the code vector and the coefficients of feedback filter in equalizer to generate the intra-block interference, and to add the intra-block interference to the code vector. Especially, the intra-block interference compensator gets all or especially 256 codewords from NR code look-up-table.

Further, there is provided an adaptive decision feedback equalizer combining with code decoder system comprising an adaptive feedback equalizer, an interference canceller, and such a code decoder in such an apparatus. Especially, the adaptive decision feedback equalizer is adapted to cancel multi-path echoes and to feed equalized symbols to the interference canceller, wherein the interference canceller is adapted to generate an inter-block interference cancelled

code vector and intra-block interference compensated code vectors and to feed them to the code decoder, and wherein the code decoder is adapted to generate the decoded code vector and soft log-likelihood ratio value and to feed the decoded code vector to interference canceller and the feedback filter part of equalizer and to feed the soft log- likelihood ratio to a forward error correction decoder.

It is preferred, when in such apparatus or in such adaptive decision feedback equalizer the code decoder is adapted as a Nordstroπi-Robinson code decoder choosing corresponding Nordstrom-Robinson code codeword as the code and outputting a decision Nordstrom-Robinson code vector as the code vector.

Furthermore, it there is preferred a method for decoding Nordstrom-Robinson code coded information, wherein a interblock interference cancelled vector and intra-block interference compensated Nordstrom-Robinson code vector are processed to calculate an Euclidean distance between the two vectors, a minimum distance and its corresponding Nordstrom- Robinson code codeword are processed and a decision Nordstrom-Robinson code vector is outputted basing on the Euclidean distances, and a log-likelihood ratio is calculated basing on the Euclidean distances. Especially, such method is adapted for execution in such apparatus or in such adaptive decision feedback equalizer.

Description of Drawings

An embodiment will be disclosed in more details with respect to enclosed drawing. There are shown in:

Fig. 1 components of a NR decoder combined with equalizer and

FEC decoder,

Fig. 2 components of a inter-block interference cancellation,

Fig. 3 components of a intra-block interference compensation,

Fig. 4 a comparison of three methods, and

Fig. 5 components of a general NR decoder combined with equalizer.

Mode for Invention

According to preferred embodiment there is provided a novel device or apparatus and method to combine NR code with time- domain equalizer for single carrier mode in Chinese

Terrestrial DTV standard. Individual or all blocks might be adapted as software, hardware or combined soft- and hardware. Especially, it comprises an inter-block interference cancellation unit 11, an intra-block interference cancellation unit 12, a NR decoder 11 as shown in Fig. 1 to provide a complexity reduced NR decoder 11.

There is executed an approximate method to calculate the Euclidean distance for complexity reduction and a method to calculate soft metric for FEC module by max-log-MAP algorithm.

Fig. 1 shows a general structure of a NR decoder 11 combined with an equalizer and a FEC decoder 7 (FEC: forward error correction) . As far as prior art components and functions

can be used they are not described in details. An input signal is or data, especially a received vector is inputted into a feed forward filter (FFF) 1. Data outputted out of the feed forward filter 1 are inputted into an input of an adder 2. Further, data outputted out of a feedback filter

(FBF) 5 are inputted into a further input of the adder. Data added in adder 2 are outputted into an interference canceller 10 and into an error generator 3. Data provided by the interference canceller 10 comprising the inter-block interference cancellation unit 11 and the intra-block interference cancellation unit 12 are inputted into the NR decoder 6. The NR decoder 6 takes the signals, especially symbols and decodes the received vector. Then, the NR decoder 6 outputs the results to the feedback filter 5, to the error generator 3, and to the forward error correction decoder 7. Further, a tap coefficient updater 4 is provided for updating tap coefficients for the feed forward filter 1 and the feedback filter 5.

In other words, Fig. 1 shows the structure of NR decoder 6 combined with equalizer and FEC decoder. The NR decoder consists of especially 5 units. Unit 1, adapted by the inter-block interference cancellation unit 11, gets soft symbols and FBF tap coefficients from equalizer, and then eliminates the interference caused by previous NR blocks with respect to inter-block interference. Unit 2, adapted by intra-block interference cancellation unit 12, compensates the interference caused by the other symbols within the same NR block, which is called intra-block interference. Unit 3 adapted by a distance calculator 13 calculates the Euclidean distance of the received vector to each codeword in a NR table. Unit 4 adapted by a decoder 14 selects the minimum distance, and then, gets its corresponding codeword. Unit 5 adapted by a soft information calculator 15 calculates the soft information.

Soft symbol outputted out of adder 2 is inputted into the inter-block interference cancellation unit 11. Symbol outputted out of the inter-block interference cancellation unit 11 is inputted into the distance calculator 13.

Further, tap coefficients of tap coefficient calculator 4 are inputted into the intra-block interference cancellation unit 12. Symbol outputted out of the intra-block interference cancellation unit 12 is inputted into the distance calculator 13. Result of the distance calculator 13 is inputted into the decoder 14 and into the soft information calculator 15. Data or symbol outputted out of the decoder 14 are inputted as decision symbol ds into the error generator 3 and into the feedback filter 5. Data or symbol outputted out of the soft information calculator 15 are outputted as LLR (log likelihood ratio) to the FEC decoder 7.

The structure of the unit 1 for the inter-block interference cancellation is shown in Fig. 2. As the decoding procedure of NR code is block based having e.g. 8 symbols in QPSK modulation, there are k symbols delay while decoding. Thus the interference caused by previous (k + 8) symbols cannot be compensated in time by FBF part in the equalizer, as mentioned in Section 2. This unit 1 stores the previous k decoded symbols and tap coefficients of feed back filter 5, then convolutes them and the output is the inter-block interference. Finally, the inter-block interference is eliminated in received blocks.

Especially, there is provided a buffer 20 for previous symbols and a buffer 21 for TAP coefficients, data of which are inputted into a convolution unit 22. After4 convolution data are inputted into an adder 23. In addition, received block from a further component or input are inputted into

adder 23 for adding and outputting data, symbol or vector X.

The detail of unit 2 for intra-block interference compensation is shown in Fig. 3. As it is impossible for the current received blocks to cancel the interference, which is caused by inter-symbols before this codeword is known, there is an alternative way to cancel the interference.

Data, especially code outputted out of a NR Code table 25 is inputted into a component for an i-th codeword N(i) . Its output is inputted into a convolution unit 28 and into an adder 29. Result of convolution is subtracted in adder 29 to output data, symbol or vector S(i) .

There are totally 256 codewords in NR code set, the inter- symbol interference for each code vector can be calculated by linear convolution between code vector and estimated channel response, which comes from equalizer. Then the inter-symbol interference is subtracted from original blocks. So, 256 "dirty" code vectors S(i), which contain only intra-block interference are got.

Then the maximum likelihood algorithm is chosen in order to achieve a better decoding gain. In unit 3, the Euclidean distances are calculated between results of adders 23 and 29, i.e. between symbol X and symbol S(i) . And the calculation for Euclidean distance is resource consuming for it contains a square operation which need lots of multiplications.

In order to reduce the complexity, it is worthy of substituting the square value calculation by other form calculations, although it may bring some system performance degradation. To simplify the calculation, absolute value can

be used to replace the square. Or more complicated functions can be used, for instance, curve approximation.

Fig. 4 shows the comparison of the three methods, and the simulation is performed in AWGN channel. It shows that the approximation brings only a little degradation.

In unit 4, the decoding for NR is operated. The decoder 14 collects all 256 Euclidean distances and finds out the minimum one and the corresponding NR codeword. The decoded block having especially 16 bits is mapped into symbols and then outputted to the interference canceller 10, the error generator 3 and the feedback filter 5.

In unit 5, soft information LLR (log-likelihood ratio) is calculated in order to help FEC module in forward error correction decoder 7 performing a soft -input iterative decoding. So, a better performance can be achieved. Unit 5 gets especially 256 Euclidean distances from unit 4 for one NR block and outputs the log-likelihood ratio LLR for forward error correction as an outer code.

As mentioned, in the Chinese patent "A time-domain equalizer combined with NR code", CN patent application number 200510026192.4, a log-MAP algorithm which is described in Section 3 is proposed to generate the soft information for FEC decoder. But the drawback of this algorithm is that the complexity is very high for implementation.

Herein an approximate max- log-MAP structure is proposed. The algorithm is described as

/// ^■ (*)«{ min (Ip-J ₁ -(I)I ² )- min Qy-S ₁ (O)F)

The max-log-MAP algorithm is still complex sometimes, for it

contains two squares for calculating the Euclidean distance in one calculation. Fortunately, the square value can be replaced by absolute value or curve approximation function. The absolute value approximated algorithm can be given as