WO2005062473A1

WO2005062473A1 - Turbo decoding with iterative estimation of channel parameters

Info

Publication number: WO2005062473A1
Application number: PCT/NL2003/000924
Authority: WO
Inventors: Dignus-Jan Moelker; Jan Stemerdink
Original assignee: Telefonaktiebolaget LM Ericsson AB
Current assignee: Telefonaktiebolaget LM Ericsson AB
Priority date: 2003-12-23
Filing date: 2003-12-23
Publication date: 2005-07-07
Anticipated expiration: 2006-06-23
Also published as: EP1700380A1; US20070286292A1; ES2283853T3; AU2003295272A1; DE60312906T2; US7840884B2; ATE358356T1; DE60312906D1; EP1700380B1

Abstract

Method and decoding device for decoding a convolutionally coded input data signal y. The input data signal is multiplied with a scaling factor Lc (8) and then demultiplexed (6). The demultiplexed input data signal LcS is then turbo decoded (5) in order to obtain decoder output likelihood ratio data Λ. The scaling factor Lc is updated (7) for a next iteration in dependence on a combination of a posteriori likelihood data based on turbo decoded output data Λ and a priori likelihood data based on the demultiplexed signal LcS.

Description

URBO DECODING ITH ITERATIVE ESTIMATION OF CHANNEL PARAMETERS

Field of the invention The present invention relates to a method for decoding a convolutionally coded input data signal y comprising multiplying the input data signal with a scaling factor

L_c , demultiplexing the multiplied input data signal L_cy , e.g. in parity signals and a systematic signal, and turbo decoding the demultiplexed input data signal Lβ in order to obtain decoder output likelihood ratio data. In a further aspect, the present invention relates to a decoder device for decoding a convolutionally coded input data signal y comprising a multiplication element for multiplying a received input data signal y with a scaling factor L_c , a demultiplexer for demultiplexing the multiplied input data signal L_cy , e.g. in parity signals and a systematic signal, and a turbo decoder for decoding the demultiplexed input data signal Lβ in order to obtain decoder output likelihood ratio data.

Prior art Such a method of decoding data and a decoder device are known from American patent publication US-B-6,393,076, which describes a method for decoding turbo codes using data scaling. Convolutionally coded input data is decoded in a near ideal manner. A portion of the input data is buffered, after which a mean of the data in the portion of the input data is calculated. Then, a root-mean-square value of the portion is calculated using the mean. A scaling factor is derived from the root-mean-square value, which scaling factor is used to scale the portion of the input data before the turbo decoding step. This disclosed method has the disadvantage that the scaling factor computation is not based on a-posteriori likelihood information, which leads to additional loss. Especially in mobile applications using these kind of coding, every additional loss has a negative effect on system performance. The publication of El-Gamal, H., High capacity Synchronous FH/SSMA networks with turbo coding, IEEE Intern. Symp. on Spread Spectrum Techniques, 1998, p973-977, provides a similar method for scaling the input data. The method consists of computing the noise variation, which is subsequently applied to form log- likelihood values. This method has the disadvantage that it assumes the signal amplitude as normalized to unity. This is a simplification not valid for practical receiving equipment, and therefore negatively impacts the system performance.

Summary of the invention The present invention seeks to provide an improved decoding scheme for use in transceivers using a maximum-a-posteriori (MAP) decoder, or related techniques such as Logarithmic MAP (LOGMAP). According to the present invention, a method according to the preamble defined above is provided, in which the scaling factor L_c is updated for a next iteration in dependence on a combination of a posteriori likelihood data based on turbo decoded output data A and a priori likelihood data based on the demultiplexed signal Lβ . By combining these a posteriori and a priori likelihood data, it is possible to improve the performance of the decoding method without the need of much additional hardware or software resources. In an embodiment of the present invention the scaling factor is updated using an estimate of the mean value of the signal amplitude c and an estimate of the noise variation

. The update of the scaling factor can e.g. be calculated according to ^Λ

L_c = _n • L_c , in which L_c is the updated scaling factor. The estimate of the mean A-²

value of the signal amplitude may be equal to c = • L_c - s_l ,

where N is the number of bits in a coding block of the input data signal, s_ is the i"' systematic bit, c is the estimation of the amplitude of the scaled systematic bits L_c • s, and Λ, is the log-likelihood ratio resulting from the most recent turbo decoder iteration. In a further embodiment, the noise variance estimation σ).. may equal ι N-l

A-² _ y_j(s: -ϊ)X p i) + (s_l' + l)X p θ) -K N -l ι=0 the probability of the /"'bit being zero is estimated like Pr{x, = θ} = P_t (0) = — and the probability of that bit being one like Pr{ _; = l} = P_t (1) = — = 1 - P_l (0) ; 1 + e ' L - s the normalised systematic bits s'_ are calculated as s[ = ^C ' ;

I N-l and where K is a bias correction computed as K = — ^ 2 • ( ₍ (0) - _P (1)) • s' - 2. N ι=0 This makes the present method accurate and fast, so that it secures the advantage

(simulations show the loss compared to the theoretical optimum is about 0.03dB), and is applicable in fast fading or time- variant channels. It has been shown that about 0.1 to 0.2dB in sensitivity improvement for turbo codes can be achieved, compared to prior art methods. In a further embodiment, the scaling factor L_c is initialized either as a fixed value, the result of an initial number of iterations using a known algorithm, filtering over subsequent iterations and coding blocks, or SΝR/SIR estimation at the input data signal. Initialization can thus be accomplished using very simple solutions or more complex but well known solutions. In an even further embodiment of the present invention, the method further comprises calculating the variation of the scaling factor in subsequent iterations and, when the variation after a predetermined number of iterations is above a predetermined threshold value, reverting to a different scaling factor calculation method and/or turbo decoding method. The different scaling factor calculation method may e.g. be a fixed scaling factor, a Log-MAX method or a SONA method (Soft Output Niterbi

Algorithm). In this manner, a simple divergence monitoring of the iterative method can be accomplished, using known algorithms as back-up. In a further aspect of the present invention, a decoder device is provided according to the preamble defined above, in which the decoder device further comprises an adaptive scaling element which is arranged to update the scaling factor L_c for a next iteration based on a combination of a posteriori likelihood data based on turbo decoded output data and a priori likelihood data based on the demultiplexed signal. The adaptive scaling element may be further arranged to execute the present method. Accordingly, the present decoder device provides an improved performance over prior art devices. Because the sensitivity improvement only requires a small hardware cost, the sensitivity improvement is a pure bonus. The important system aspects, which influence sensitivity, are signal strength, i.e. range/coverage/battery life, noise figure, and interference, i.e., air interface capacity. Hence, the advantage of the present invention can be translated into a coverage increase (<1% more range), increased mobile battery life time (2 to 4% less transmit power), relaxed noise figure requirements (0.2dB less), or improved capacity users on the air interface (2 to 4% more). This yields a larger cell-coverage and lower power usage for the mobile, which in turn gives less interference, and therefore more capacity. In an even further aspect, the present invention provides a computer program product, which comprises computer executable code, which when loaded on a processing system, provides the processing system with the capability to execute the present method. The processing system may comprise a microprocessor and peripheral equipment, a digital signal processor or a combination of both to execute the present method.

Short description of drawings The present invention will be discussed in more detail below, using a number of exemplary embodiments, with reference to the attached drawings, in which Fig. 1 shows a block diagram of an embodiment of the decoding scheme according to the present invention.

Detailed description of exemplary embodiments The present application may be advantageously applied in decoding of systematic forward error-correcting codes, such as the parallel concatenated convolutional turbo codes found in third generation mobile telecommunication systems (3GPP). The present invention can be employed in transceivers using a maximum-a-posteriori (MAP) decoder, or related techniques, such as Log-MAP. Iterative MAP and Log-MAP decoding are asympotically optimum. Sub-optimal decoding algorithms, such as soft output Niterbi algorithm (SONA) or approximated LOGMAP (Log-Max), are simpler implementation-wise, yet about 0.5dB short of the MAP and Log-MAP performance. The problem with achieving the full extent of MAP performance is the requirement that the input data is defined as a (log) likelihood ratio. With the present invention it is possible to scale the decoder input data in a time- variant, fading channel, quickly and accurately, in order to form (log)likelihood values. The present method uses the a-posteriori likelihood data to form probability information of the decoded bits. This is combined with the decoder a-priori likelihood data, in order to optimally determine the desired signal mean and noise variance. These are used to iteratively scale the a-priori information. The scaling factor should be initialized. The invention foresees in several different methods for this: Use the scaling factor from the previous coding block (filtered or not); - Run the first iteration(s) with SONA or Log-Max; Deploy any method as initial value setting; or Any combination of these. Fig. 1 shows a block diagram of the iterative scaling algorithm 10 for decoding convolutionally coded data. The received symbols y are multiplied by a scaling factor L_c using a multiplier 8 before they are de-multiplexed using a demultiplexer 6 into systematic bits s_t and parity bits par , par₂ that are input to a turbo decoder 5. After an iteration of the decoder 5 an adaptive sealer 7 makes an estimation of the signal amplitude and the noise variance, based on the decoder output log-likelihood ratios Λ_; .

From these estimations a new scaling factor L_c is computed for the next iteration. The turbo decoder 5 comprises as shown in Fig. 1, interleavers 13, 14, de- interleaver 15, and decoders 11, 12. More in detail, the turbo decoder 5 comprises first and second soft input soft output (SISO) decoder sections 11, 12 which receive data from the demultiplexer 6. The functioning of the turbo decoder 5 is known to the person skilled in the art (see e.g. US-B-6,393,076) and needs no fiirther explanation in this description. In the following the operation of the adaptive sealer 7 is explained in detail. The calculations performed by the adaptive sealer 7 are described as algorithms, calculations and the like. For the person skilled in the art, it will be clear that the algorithms and calculations as performed by the various blocks and elements of the decoder algorithm 10 may be implemented using software, hardware resources or a combination of both, such as analogue electronics, logical circuitry, signal processors, etc. The signal amplitude is computed in block 20 from the log-likelihood ratios Λ₍ resulting from the most recent turbo decoder 5 iteration, and from the systematic bits as follows:

where N is the number of bits in a coding block and s, is the i^th systematic bit. c is the estimation of the amplitude of the scaled systematic bits L_c - s_t ; if E_s is the

estimated symbol energy of the signal component of input sequence s. , then ^E_s is the estimated amplitude of this signal component; then we find that

In block 21 the estimated amplitude c is used to normalize the systematic bits s[ giving:

From the log-likelihood ratios Λ, an estimation of the logical bit probabilities can be computed in block 22; the probability of the f' bit being 0 is estimated like

Pr{*, = 0} = P_l(0) =_ϊ-i^

and the probability of that bit being 1 like

Pr{x_l = l} = P_ι ) = -^_ζ- = l- P_l(0) 1 + e ' The non-linear relationship between Λ_; and logical bit probabilities Pi(0) and

Pi(l) can be computed or calculated based on a look-up-table. These probabilities together with the normalised systematic bits s'_ can be used to

of the variance of the noise in block 23, according to a = -^— _J(_S -l)X P_[(l) + (_Sl' +l)X P_lφ) - K jy ~ * (=o where K is a bias correction computed as follows:

Then the relation between σ_n, and E_s and N₀ is given by

When the above equations are rewritten to

and use c = L ^■ JE. <JE = L„ the new estimated optimum scaling factor can be computed in block 24 as:

Summarizing, the new scaling factor in the amplitude and variance estimations, and in the current scaling factor can be expressed as follows: c - σ_n,

This calculation is performed using multiplier 25 and memory latch 26, after which the current scaling factor is applied again to the input data y. Since the turbo decoder 5 is iterative itself, merging it with another iterative algorithm (that of the adaptive sealer 7) requires careful consideration in order to avoid divergence or oscillation. Initialisation of the scaling factor L_c to a reasable starting value proves to be sufficient. In the absence of proper initialisation information, the first decoding iterations can be done with known LogMax or SONA techniques. These are not scaling factor dependent, and can therefore be used to let the a-posteriori likelihood values converge. Other techniques to acquire initialisation information are filtering over subsequent iterations and coding blocks, and SΝR/SIR estimation at the input, see for instance the method described in Summers, T.A., Wilson, S.G., SΝR Mismatch and Online Estimation in Turbo Decoding, IEEE Tr. On COM, Nol. 46, Νo.4., April 1998, pp. 421-423. Moreover, divergence can be detected by monitoring the development of the scaling factor L_c . Normally, it should converge in a few iterations, and then hardly change. In the case of detected divergence, one can fix the scaling factor L_c , or switch to Log-Max or SONA. The description of the embodiment above, and set forth in the claims, is based on the LOG-MAP type of turbo decoder. A person skilled in the art will recognize that the ideas of the disclosed method are not restricted to LOG-MAP decoding, but can also be formulated for variations on the LOG-MAP decoding method, and for MAP decoding.

Claims

1. Method for decoding a convolutionally coded input data signal y comprising

- multiplying the input data signal with a scaling factor L_c ; - demultiplexing the multiplied input data signal L_cy ;

- turbo decoding the demultiplexed input data signal Lβ in order to obtain decoder output likelihood ratio data, characterized in that, the scaling factor L_c is updated for a next iteration in dependence on a combination of a posteriori likelihood data based on turbo decoded output data A and a priori likelihood data based on the demultiplexed signal Lβ .

2. Method according to claim 1, in which the scaling factor L_c is updated using an estimate of the mean value of the signal amplitude c and an estimate of the noise variation σ .

3. Method according to claim 2, in which the scaling factor is updated according 2 to L_c = — — — • L_c , in which L_c is the updated scaling factor. c • σ„,

4. Method according to claim 2 or 3, in which the estimate of the mean value of 1 JV-1 the signal amplitude is equal to c = — ^ sgn(Λ, ) -L_c -s_l , N ;=0 where N is the number of bits in a coding block of the input data signal, s_t is the i^th systematic bit, c is the estimation of the amplitude of the scaled systematic bits L_c • s_t and Λ₍ is the log-likelihood ratio resulting from the most recent turbo decoder iteration.

5. Method according to claim 2, 3 or 4, in which the noise variance estimation ² equals <τ,², - - -∑tø' - l)² . pχi) + (s_l' ₊ ϊ)^{2 ■} P,(0) - K ; jy — i- ;=o 1 the probability of the i^th bit being zero is estimated like Prix, = θ} = P_t (0) = — - 1 + e '

and the probability of that bit being one like Pr{x₍ = l} = P_t (1) = _j- = 1 - P_t (0) ; 1 + e ' L - s the normalised systematic bits s_t' are calculated as s[ = ^c „ ' ;

I N-l and where .ST is a bias correction computed as K - — ^ 2 • (P_l (0) - P_l (1)) - $ - 2. N (=0

6. Method according to one of the proceeding claims, in which the scaling factor L_c is initialized either as a fixed value, the result of an initial number of iterations using a known algorithm, filtering over subsequent iterations and coding blocks, or SΝR/SIR estimation at the input data signal y .

1. Method according to one of the proceeding claims, further comprising - calculating the variation of the scaling factor in subsequent iterations and, when the variation after a predetermined number of iterations is above a predetermined threshold value, reverting to a different scaling factor calculation method and/or turbo decoding method.

8. Decoder device for decoding a convolutionally coded input data signal y comprising

- a multiplication element (8) for multiplying a received input data signal y with a scaling factor L_c ;

- a demultiplexer (6) for demultiplexing the multiplied input data signal L_cy ; - a turbo decoder (5) for decoding the demultiplexed input data signal in order to obtain decoder output likelihood ratio data A , characterized in that, the decoder device (10) further comprises an adaptive scaling element (7) which is arranged to update the scaling factor L_c for a next iteration based on a combination of a posteriori likelihood data based on turbo decoded output data ^"A and a priori likelihood data based on the demultiplexed signal Lβ .

9. Decoder device according to claim 8, in which the adaptive scaling element (7) is further arranged to execute the method according to one of the claims 2 through 7.

10. Computer program product, which comprises computer executable code, which when loaded on a processing system, provides the processing system with the capability to execute the method according to one of the claims 1 through 7.