KR100362851B1 - apparatus for turbo code decoder and estimate method for channel state of the same - Google Patents

Abstract

The present invention provides a channel state estimation method for improving the decoding performance of a turbo code. In the flat Rayleigh fading channel environment having a correlation, a soft output of a turbo code is used to estimate a channel fading size And provides a more optimized channel estimation scheme for estimating the noise variance. The soft output value of the turbo code is fed back to the input of the iterative channel estimator to update the tap coefficient of the iterative channel estimator so that the fading size and noise variance of the channel approaches the actual value as the iteration number increases . Therefore, since the channel state required for the decoding process of the turbo code is less error-estimated, the performance of the turbo code can be improved.

Description

A turbo code decoding apparatus and a channel state estimation method therefor,

The present invention relates to decoding of a turbo code, and more particularly, to a device for turbo code decoding and a channel state estimation method thereof.

Recently, turbo codes have been studied as error correction codes of next generation mobile communication.

It is known that the performance of this turbo code is better than that of convolutional codes, which are currently used in mobile communication environments, and exhibit a BER (Bit Error Rate) performance close to the Shannon limit in the Gaussian channel.

The basic structure of such a turbo code is shown in Fig.

As shown in FIG. 1, two simple RSC (Recursive Systematic Convolutional) encoders for generating parity symbols using data d _k composed of a plurality of information bit frames are connected in parallel.

That is, the turbo encoder outputs the information bit 1 as one X _k and the information bit 1 through the first RSC encoder 2 to obtain Y _1k (3).

A signal 9 obtained by passing the information signal 1 through an interleaver 6 having the same size as a frame of a plurality of information bits is passed through another second RSC encoder 5 to generate Y _2k 7).

Therefore, the output of the turbo code has the dual parity information as the output of the second RSC encoder using the output of the first RSC encoder as well as the output of the interleaver.

In order to obtain the desired coding rate in the turbo encoder, the output signal is punctured through the puncturer 4 to Y _1k (3) and Y _2k (7).

The parity bit Y _k (105), which is finally obtained, is transmitted together with X _k to the receiving end.

The receiving end decodes the turbo code-encoded codewords by coupling two decoders of the structure shown in FIG. 2 in series and decodes them.

Each of these decoders has an input having a value other than 0 or 1, i. E., A soft input, or an output having a value other than 0 or 1, i.e., a soft output.

Fig. Referring to the decoder through a 2, when, first signals via a code word channel of the turbo encoder as shown in Fig. 1, the X _k and Y _k, X _k is the signal information bit is passed through the channel Y _k parity bits The signal passed through the channel.

The parity bit Y _1k corresponding to the first RSC encoder 2 through the MUX to the received parity signal Y _k is transmitted to the first decoder 10 and the parity corresponding to the second RSC encoder 5 The bit Y _2k is sent to the second decoder 12.

First, the first decoder 10 decodes using X _k and Y _1k , and then interleaves through the interleaver 11.

Then, the second decoder 12 decodes the signal using the interleaved signal and the parity bit Y _2k .

At this time, if it is not desired to perform iterative decoding, the output signal of the second decoder 12 is deinterleaved through the deinterleaver 13, and then the signal is determined through the hard decision unit 14.

However, in order to perform iterative decoding, the output signal of the second decoder 12 is deinterleaved and decoded by the first decoder 10 using the received signals X _k and Y _k .

This iterative decoding can be performed until the desired performance is obtained.

The turbo decoding method generally uses MAP (Maximum A Posteriori) decoding or SOVA (Soft Output Viterbi Algorithm) decoding.

It is known that the performance of the dual decoder is better than that of the MAP decoding scheme, and the MAP decoding scheme is known to be very sensitive to the channel state in the fading channel environment.

Recently, many researches have been made on the performance analysis of turbo codes in the flat Rayleigh fading channel environment. However, most of these studies are based on the assumption that the fading magnitudes are independent of each other by the channel interleaver, Assume that you know the variance and fading size.

However, in actual systems, it is very difficult to completely estimate these values because the channel state must be estimated at the receiving end. Therefore, the performance of turbo codes is reduced in the fading environment.

Channel estimates for turbo codes have been discussed by M. A. Jordan and R. A. Nichols and have been studied extensively for BSC (Binary Symmetric Channels) and AWGN channels.

According to this paper, turbo coding does not affect the performance even if the noise variance is estimated to be less than 3 dB in the AWGN channel, but it shows that when the noise variance is estimated to be more than 3 dB, the performance decreases sharply.

M. C. Valenti has proposed a simple channel estimation method for turbo codes in flat fading channels and analyzed the effects of noise variance and fading size estimation error on turbo codes.

As a result of this analysis, it is analyzed that the error of fading size estimation has more influence on performance than turbo code.

As described above, the present invention provides a channel state estimation method for accurately estimating a channel state in a flat Rayleigh fading channel environment in order to improve a decoding performance of a turbo code.

1 is a diagram showing a structure of a turbo encoder according to the prior art;

2 is a diagram illustrating a structure of a turbo code decoder according to the prior art;

3 shows a channel state estimation apparatus for turbo code decoding according to the present invention.

4 is a diagram illustrating a repetitive channel estimator according to the present invention;

5 is a graph illustrating BER performance of a turbo code using the channel estimation method according to the present invention.

6 is a graph showing the comparison between the actual channel noise analysis and the estimated noise variance according to the present invention

Figure 7 is a graph = 3dB shows the fading size estimated by the proposed iterative channel estimator for each repetition of the actual fading size and turbo decoding

Fig. 8 is a flowchart = 3dB shows the MSE for the actual fading size according to the repetition times of the turbo decoding and the fading size estimated by the proposed iterative channel estimator

DESCRIPTION OF THE REFERENCE NUMERALS

100: Transmission unit 110: Turbo code unit

120: channel interleaver 130: BPSK modulator

200: Receiver 210: BPSK demodulator

220: channel deinterleaver 230: turbo decoder unit

240: Repetition channel estimation unit 241: Information bit

242, 243, 300, 400: coupler

According to an aspect of the present invention, there is provided an apparatus for decoding a turbo code, the apparatus comprising: a transmitter for encoding, modulating and transmitting an information signal including a plurality of information bit frames; A receiver for recovering a received signal in combination with a fade signal and an AWGN signal while passing through a correlated flat Rayleigh fading channel, the receiver comprising: a BPSK demodulator for demodulating a received signal; and the demodulated signal deinterleaving the deinterleaver for, receiving the signal demodulated by the BPSK demodulator soft to obtain the d _k values estimated using the output value of LLR values, necessary for decoding through a weighting coefficient is repeatedly updated as An iterative channel estimator for estimating a channel state; There is composed by including a turbo decoder for decoding the encoded codeword output from the channel interleaver in the turbo code.

According to another aspect of the present invention, there is provided a channel state estimation method for turbo code decoding, the method comprising: combining a fading signal and an AWGN signal while passing through a flat Rayleigh fading channel, A method for estimating a channel condition for recovering a received signal, the method comprising: generating an information bit generated by combining the received signal with an initial channel estimate generated by a training sequence; Estimating a fade signal by combining a weight value of each of the plurality of fade signals with a weight value of each of the plurality of fade signals; combining the estimated fade signal with the generated information bits to obtain a noise variance value; And decoding the received signal.

Other objects, features and advantages of the present invention will become apparent from the detailed description of the embodiments with reference to the accompanying drawings.

A preferred embodiment of a channel state estimation method for turbo code decoding according to the present invention will be described with reference to the accompanying drawings.

FIG. 3 is a diagram illustrating a channel state estimation apparatus for turbo code decoding according to the present invention.

Referring to FIG. 3, an information sequence b _i (1? _I ? L) composed of a plurality of information bit frames is encoded by a turbo coding unit 110 having a code rate Rc and a random interleaver.

Then, the encoded sequence x _{k (1} ? _K? M (= L / R _C )) is interleaved through the channel interleaver 120 and then modulated through a BPSK (Binary Phase Shift Keying)

The modulated signal d _k in the BPSK modulator 130 is then transmitted to the receiver via the correlated flat Rayleigh fading channel.

The received signal r _k is combined with the fading signal a _k and the AWGN signal n _k through the correlated flat Rayleigh fading channel.

This can be expressed by the following equation (1).

r _k = a _k d _k + n _k

In Equation (1), the fade signal a _k is a sample of a complex Gaussian random process having an average of zero correlation.

And the AWGN signal n _k has an average of 0, to be.

Where E _b is the energy per bit and N ₀ is the power spectral density of the white noise in one frequency component only.

This fading channel can be expressed by the autocorrelation function of the fading process a _k as a jakes model as shown in the following equation (2).

In Equation (2), J ₀ (·) is a zero-order Bessel function, f _d is a maximum Doppler frequency, T _s is a symbol period, and * is a complex conjugate.

When thus via the BPSK flat Rayleigh fading channel is a modulated signal d _k by the modulator 130 with a correlation with the fade signal a _k and the AWGN signal n r _k signal _k it is bonded transmitted to the receiving end, the receiving end ( 200 first depunctures the received r _k signal.

Demodulator 210, and then deinterleaved through a channel deinterleaver 220. The demodulator 210 demultiplexes the demodulated signal through a channel deinterleaver 220,

At the same time as deinterleaving, the iterative channel estimator 240 receives the demodulated signal through a BPSK demodulator 210, estimates a channel state required for decoding through repetition, and inputs the demodulated signal to the turbo decoder 230.

The turbo decoder 230 decodes the turbo code-encoded codewords input from the channel deinterleaver 220 based on the channel state estimated by the iterative channel estimator 240, do.

The channel states estimated by the repeated channel estimator 240 will be described in more detail with reference to the drawings.

4 is a diagram illustrating a repetitive channel estimator according to the present invention.

First, r _k (= a _k d _k + n _k ) configured as shown in Equation (1) is received at the receiving end.

Then, it is initially assumed that we know all about l <k by the training sequence, and we define it as information bit y _l = r _l d _l .

At this time, the training sequence knows d _l because it sends promised information to each other. At this time, d ₁ has a value of +1 or -1.

Therefore, multiplying the received r _l by the estimated value d _l for the transmitted data signal yields the information bit y _l (= a _l + n _l d _l ).

At this time, the initial estimated value of the transmitted data signal d _k is obtained by simply taking the sign of the received r _k as shown in the following equation (3).

At this time, a Minimum Mean-Squared Error (MMSE) scheme is used to minimize a channel estimation error of the estimated value obtained from Equation (3).

The error signal e _k is expressed by the following equation (4).

In Equation (4)

ego, H and T denote Hermitian and Transpose, respectively.

And a cost function for minimizing the error signal in the MSE (Minimum Squared Error) is expressed by Equation (5).

If a gradient is taken as the cost function shown in Equation (5), the following Equation (6) is obtained.

Then, the optimal weight vector for which the value of Equation (6) is set to 0 is expressed by Equation (7).

Here, A and c are respectively expressed by the following equations (8) and (9).

In Equations (8) and (9), N is a filter coefficient.

If the data signals d _k are independent from each other, the elements of the autocorrelation matrix A And the element of the cross-correlation vector c Can be expressed by the following equations (10) and (11), respectively.

, 0 = j < = N-1

In Equation (10) and Equation (11) Estimated from the formula for Can be expressed by Equation (12). &Quot; (12) &quot;

= 1-2P

In this case, P in Equation (12) Is a probability of being erroneously estimated, and Equation (12) Wow Is represented by Equation (13). &Quot; (13) &quot;

Then, by applying Bayes' theorem to Equation (13), the following Equation (14) is obtained.

In Equation (14) Wow Is equal to the probability 1-P that no error will occur, respectively.

And Wow Is equal to the probability P that the error occurs.

therefore = The equation (14) is the result of the equation (12).

And the soft-output value LLR (Log-Likelihood Ratio) of the turbo code is expressed by the following equation (15).

At this time, the right side of Equation (14) uses the LLR value shown in Equation (15) Can be approximated by an average reliability to be determined as 1 and -1, using Equation (16) and Equation (17) as follows.

Thus, Equation (14) is approximated as Equation (18).

Substituting Equation (18) into Equation (10) and Equation (11) approximates Equation (19) and Equation (20) as follows.

Therefore, the approximated optimal weight vector for the equation (17) is expressed by the following equation (21).

In Equation 21, to be.

If f _d T _s << 1 and N << The backward weighting coefficient w _k is _calculated for all k .

therefore Lt; RTI ID = 0.0 &gt; a &lt; / RTI &gt; .

The initial weight Is an approximate value of the weight coefficient value w _k , the error can be larger than the correct value. At this time, the weight coefficient value w _k is estimated to be accurate So that the weight coefficient value w _k must be obtained more accurately.

Thus, by using the accurate value of the weight coefficient value w _k , You have to find the value, The value of d _k ( ) Value of the a _k part of y _k (= a _k + n _k d _k ) according to how accurate the value is.

At this time, Value can be obtained the most accurate value when the decoding process of the turbo code is finished.

Therefore, the value of d _k obtained in the decoding process of the turbo code needs to be calculated .

Using Equation (21), the value of w _k is approximated And more accurate values are obtained.

Therefore, w _k value can be accurately estimated becomes closer to a _k value to the actual value.

To obtain the AWGN signal n _k , the following calculation process is performed.

First, y _k and the estimated a _k ( Through an adder 242. [

At this time It is negative (-) to have a value, y _k has a value of positive (+) also calculated by substituting y _k = a _k + n _k d _k Since y _k, as defined in the above description and the following equation (22) Together.

a _k + n _k d _k - a _k n _k d _k

In Equation (22), d _k is +1 or -1,

y _k + n _k .

In this case, the sign of n _k is positive (+) or negative (-), but even if n _k itself is random noise, its sign is also random noise.

In this way distributed to the n _k output via the adder 242 is obtained from the mean minus the adder 243 of the n _k n at _k, and then taking the square on the obtained value is obtained by averaging.

And a _k ( _k ) estimated as a fade signal obtained above together with r _k output from the channel deinterleaver 220 ) And the AWGN signal n _k , ( ) To the turbo decoder unit 230. [

During subsequent repetitions of turbo decoding The noise variance required for the value calculation uses the noise variance estimated in the previous iteration.

Then, the weight coefficient w _k for channel state estimation according to the repetition can be obtained by Equation (21).

As described above, a performance analysis performed on a correlated Rayleigh fading channel using a turbo code having a repetitive channel estimator according to the present invention is as follows.

For simulations, we first used a finite impulse response (FIR) filter with a fading rate of f _d T _s = 0.005 and a repetitive channel estimator with a tap coefficient of 32, a 25 x 36 block channel interleaver, and a code rate R c = , And the code generator And a turbo code having a random interleaver with L = 450 is used.

In addition, a log-MAP scheme is used in which the decoding algorithm is repeated five times.

A comparison of the simulated performance through the channel state is shown in FIG.

FIG. 5 compares the BER performances of the turbo codes according to the channel state estimation using the proposed channel estimation method and the non-iterative channel estimation method when the channel information is known accurately.

It can be seen that the iterative channel estimation scheme proposed in the BER of the non-repetitive channel estimation scheme is improved by about 0.25 dB.

In addition, when the channel state is known accurately and compared with the proposed iterative channel estimation method, the difference is about 0.3 dB.

The actual channel noise analysis and the estimated noise variance are compared and shown in FIG.

6, The proposed iterative channel estimation method converges to the actual noise variance while the non-iterative channel estimation method estimates the actual variance value at over 5.5dB.

Therefore, it can be seen that the proposed method shows excellent performance.

Figure 7 = 3dB shows the fading size estimated by the proposed iterative channel estimator for each repetition of the actual fading size and turbo decoding.

8, = 3dB shows the MSE for the actual fading size according to the repetition times of the turbo decoding and the fading size estimated by the proposed iterative channel estimator.

Referring to FIG. 8, it can be seen that the estimated fading size approaches the actual fading size as the number of repetitions of turbo decoding increases.

That is, since the channel state is estimated at every repetition in the turbo decoding, Is more accurately estimated for each iteration.

The estimated W _k is the soft-output value is improved by using the weight of LLR values (soft-output) is reduced, an error of the estimated fading size.

As described above, according to the channel state estimation method for turbo code decoding according to the present invention, as the number of iterations increases, the fading size and the noise variance are further improved, and the BER performance of the turbo code approaches the BER performance when the channel state is known accurately .

As described above, the apparatus and the channel state estimation method for turbo code decoding according to the present invention can improve reliability in turbo code decoding by more accurately estimating channel states for turbo codes in a flat Rayleigh fading channel environment.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit of the invention.

Therefore, the technical scope of the present invention should not be limited to the contents described in the embodiments but should be determined according to the claims.

Claims (8)

A transmission unit for encoding and modulating an information signal composed of a plurality of information bit frames and a transmission unit for combining the fade signal and the AWGN signal while passing through a flat Rayleigh fading channel having a correlation between the signals transmitted from the transmission unit, A receiver for restoring a received signal, The receiver includes a demodulator for demodulating the received signal, A deinterleaver for deinterleaving the demodulated signal in the demodulator, A repetition channel estimator for receiving a demodulated signal from the demodulator and obtaining an estimated d _k value using an LLR value as a soft output value and estimating a channel state required for decoding through a weight coefficient value that is updated repetitively; And a turbo decoder decoding the encoded codeword output from the deinterleaver based on the channel state estimated by the iterative channel estimator using a turbo code. The method according to claim 1, And the estimated d _k value is an LLR value. The method according to claim 1, Wherein the demodulator is a BPSK demodulator. A method for estimating a channel state for recovering a received signal in combination with a fade signal and an AWGN signal while passing through a flat Rayleigh fading channel in which a signal transmitted from a transmitter is correlated, Generating an information bit by combining the received signal with an initial channel estimate value generated by a training sequence, Estimating a fade signal by combining a weight value with each of the generated information bits; Combining the estimated fade signal and the generated information bit to obtain a noise variance value; And decoding the received signal based on the estimated fade signal and the noise variance. 5. The method of claim 4, Wherein the channel estimation value utilizes a channel state value obtained in a decoding process of the turbo code. 5. The method of claim 4, Wherein the channel estimation is performed by the iterative calculation. The method according to claim 6, The initial value of the weight And estimating a channel state for turbo code decoding. 5. The method of claim 4, Wherein the noise variance value is obtained by subtracting a value averaged in the estimated fade signal and squaring the obtained value, and then averaging the noise variance value.