Decoding method and system for real-time wireless channel estimation

[0001] 1. Field of the Invention

[0002] The present invention relates to the technical field of error correcting codes (ECCs) and, more particularly, to a decoding method and system for real-time wireless channel estimation.

[0003] 2. Description of Related Art

[0004] While a digital communication system transfers data, a receiving end may receive error messages because of non-predictable interferences or noises. Therefore, a channel coding is used to effectively reduce the bit error rate, without raising the transmission power, which typically occupies partial bandwidth for transmission but ensures that the bit error rate is controlled in an acceptable range. In recent years, the data transmission is required day by day. For a future data transmission, the coverage becomes wider and the transmission rate becomes faster. Accordingly, the channel coding is considered as an important part in a wireless transmission system design.

[0005] Turbo codes are an essential breakthrough in the recent channel coding, which can provide the superior error correcting capability and almost the ideal Shannon limit in transmission. Accordingly, the turbo codes are largely used in the field of wireless transmission, including satellite communications, digital image transmissions and 3GPPs.

[0006] FIG. 1 is a diagram of a typical configuration of communication system using a turbo code based on the UMTS WCDMA standard. As shown in FIG. 1, the communication system includes a turbo encoder 110, a modulator 120, a demodulator 130 and a turbo decoder 140. The turbo encoder 110 is given by an example of code rate ⅓ with 8-state turbo encoder. FIG. 2 is a block diagram of an embodiment of the turbo encoder 110. As shown in FIG. 2, the turbo encoder 110 includes an interleaver 210, a first encoder 220 and a second encoder 230. The turbo encoder 110 receives an information sequence μ, and produces a codeword c by a turbo coding. The codeword c includes systematic bits μ, first parity bits p produced by the first encoder 220, and second parity bits q produced by the second encoder 230; i.e., the codeword c={μ,p,q}.

[0007] The modulator 120 performs a BPSK modulation on the codeword c={μ,p,q} to thereby produce a modulated signal x, where xε{+1,−1}. The modulated signal x is passed through an AWGN channel for transmission. The demodulator 130 receives the modulated signal x through the AWGN channel and demodulates it to thereby produce a received signal y expressed by y_i=x_i+n=±1+n, for n denotes noises in the AWGN channel and contains zero mean and variance σ².

[0008] FIG. 3 is a block diagram of the turbo decoder 140. As shown in FIG. 3, the turbo decoder 140 includes a first decoder 310, an interleaver 320, a second decoder 330, and a deinterleaver 340. The received signal y is input to the turbo decoder 140 for turbo decoding. The log-likelihood ratios (LLRs) Λ_in of the received signal y consist of LLRs of the systematic bits μ, the first parity bits p and the second parity bits. Namely, we have Λ_in={Λ_μ, Λ_p, Λ_q}, where notation Λ_μ denotes the LLRs of the systematic bits μ, notation Λ_p denotes the LLRs of the first parity bits p, notation Λ_q denotes the LLRs of the second parity bits q, and notation Λ_in can be expressed by

[0009] ${\Lambda }_{in}=\log \frac{P\left(y_i|x_i=+1\right)}{P\left(y_i|x_i=-1\right)}=\frac{2}{{\sigma }^2}{y}_i.$

[0010] The first decoder 310 receives Λ_μ, Λ_p and a first prior information Prior₁⁽ⁱ⁾, and accordingly produces a first extrinsic information E₁⁽ⁱ⁾, for i denotes i-th iteration and Prior₁⁽⁰⁾ is initially set to zero. The interleaver 320 produces a second prior information Prior₂⁽ⁱ⁾ based on the first extrinsic information E₁⁽ⁱ⁾ and produces interleaved systematic bits μ^π having LLRs Λ_μ^π based on the systematic bits.

[0011] The second decoder 330 receives Λ_μ^π, Λ_q and the second prior information Prior₂⁽ⁱ⁾ to thereby produce a second extrinsic information E₂⁽ⁱ⁾. The second extrinsic information E₂⁽ⁱ⁾ is passed through the deinterleaver 340 to thereby produce the first prior information Prior₁⁽ⁱ⁾.

[0012] After a certain amount of iterations, the iteration can be stopped after the second decoder 330 ends the decoding, and the second decoder 330 outputs a second LLR Λ_out,2⁽ⁱ⁾=Λ_μ+E₁⁽ⁱ⁾+Prior₁⁽ⁱ⁾.

[0013] In addition, the iteration can be stopped after the first decoder 310 ends the decoding, and the first decoder 310 outputs a first Λ_out,1⁽ⁱ⁾=Λ_μ+E₁⁽ⁱ⁾+Prior₁⁽ⁱ⁻¹⁾.

[0014] In wireless communication systems, a signal power can be adjusted by a power control mechanism, and each code block has a different channel noise power. The Log-MAX decoding procedure depends on the channel characteristics for optimization, which involves a non-linear operation expressed by log(e^x+e^y)=max(x,y)+log(1+e^−|x−y|).

[0015] The Max-Log-MAP decoding procedure is a simplified version of the Log-MAX decoding procedure, which eliminates the nonlinear item log(1+e^−|x−y|). Therefore, in the Max-Log-MAP decoding procedure, the LLRs can be set directly to values to be received, i.e., Λ_in=y_i, without performing a channel estimation, but the performance is reduced by about 0.5 dB than the Log-MAX decoding procedure. When the channel estimation is not accurate, the Log-MAX decoding performance is reduced significantly in comparison with the Max-Log-MAP decoding performance.

[0016] In U.S. Pat. No. 6,393,257 granted to Holtzman for a “Wireless communications receiver and decoder for receiving encoded transmissions, such as transmissions using turbo codes, and estimating channel conditions”, channel estimation is achieved by computing the ensemble average E[|y|] of first-order absolute values of a received signal y and the ensemble average of the square received signal y E[y²]. However, computing y² consumes much computational capability. For example, if the received signal y has K message bits, the square E[y²] for the received signal y can be expressed by:

[0017] $E\left[y^2\right]=\frac{{\left(y_0\right)}^2+{\left(y_1\right)}^2+\dots +{\left(y_{K-1}\right)}^2}{K}.$
From this equation, it is known that y_i² is computed K times for finding E[y²], which is impractical.

[0018] In addition, an improved channel estimation method is proposed in the article “Effect of Mismatched SNR on the Performance of Log-MAP Turbo Decoder,” IEEE Trans. on Vehicular Technology, vol. 52, Issue 5, September 2003, which firstly performs the Max-Log-MAP decoding one time, then uses the hard decision based on a temporary LRR to estimate the channel signal-to-noise ration (SNR), and finally performs the Log-MAX decoding. In this case, since the Max-Log-MAP decoding is performed only one time, computing the square E[y²] for the received signal y is required in the Log-MAX decoding.

[0019] Therefore, it is desirable to provide an improved turbo decoding method system to mitigate and/or obviate the aforementioned problems.

[0020] An object of the present invention is to provide a decoding method and system for real-time wireless channel estimation, which can understand the channel condition (Â, {circumflex over (σ)}) when perform a Log-MAP decoding procedure is performed and therefore avoid the problems of incorrect channel estimation and low Log-MAX decoding performance.

[0021] In accordance with one aspect of the invention, there is provided a decoding method for real-time wireless channel estimation, which decodes a received signal with a noise. The received signal is coded by a concatenation code. The method includes the steps: (A) receiving the received signal with K message bits transmitted between a base station and corresponding stations; (B) applying a Max-Log-MAP decoding procedure to perform a certain amount of decoding on the received signal; (C) using temporary log-likelihood ratios (LLRs) to compute hard decisions of the K message bits of the received signal; (D) computing an ensemble average of the K message bits based on the hard decisions and the K message bits, and assigning the ensemble average to an estimated amplitude of the received signal; (E) computing an ensemble average of a first momentum based on absolute values of the K message bits; and (F) computing a ratio based on the ensemble average of the first momentum and the estimated amplitude of the received signal, and computing an estimated variance of the noise based on the ratio.

[0022] In accordance with another aspect of the invention, there is provided a decoding system for real-time wireless channel estimation, which decodes a received signal with a noise. The received signal is coded by a concatenation code and contains K message bits. The K message bits are grouped into systematic bits, first parity bits and second parity bits. The system includes a first decoder, an interleaver, a second decoder, a deinterleaver, a channel estimator and a controller. The first decoder receives a systematic bit log-likelihood ratio (LLR), a first parity bit LLR and a first prior information, and accordingly performs a Max-Log-MAP or Log-MAP decoding for producing a first extrinsic information and a first output LLR. The interleaver is connected to the first decoder in order to receive the systematic bit LLR and the first extrinsic information for re-arrangement to thereby produce an interleaved systematic bit LLR and a second prior information. The second decoder is connected to the interleaver in order to receive the interleaved systematic bit LLR, a second parity bit LLR and the second prior information, and accordingly performs the Max-Log-MAP or Log-MAP decoding for producing a second extrinsic information and a second output LLR. The deinterleaver is connected to the second decoder in order to receive the second extrinsic information for re-arrangement to thereby produce the first prior information. The channel estimator is connected to the first and the second decoders in order to perform a channel estimation in accordance with the received signal, the first output LLR and the second output LLR to thereby produce an estimated amplitude of the received signal and an estimated variance to the first decode and the second decoder. The controller is connected to the channel estimator, the first decoder and the second decoder in order to control the first and the second decoders to first perform a certain amount of decoding on the received signal by a Max-Log-MAP decoding procedure and then by a Log-MAP decoding procedure. The first decoder and the second decoder adjust the received signal in accordance with the estimated amplitude and variance when the Log-MAP decoding procedure is performed, and therefore produce an adjusted received signal.

[0023] The decoding system further includes a memory connected to the controller and storing a specific table. The controller is based on a code rate of the concatenation code and the K value to look up the specific table and thereby obtain the certain amount.

[0024] In addition, the first decoder and the second decoder adjust the first second extrinsic information and the second extrinsic information in accordance with the estimated amplitude and variance when the Log-MAP decoding procedure is performed, and therefore produce a scaled first extrinsic information and a scaled second extrinsic information respectively.

[0025] The adjusted received signal is obtained by dividing the received signal by

[0026] $\frac{\hat{A}{\sigma }^2}{2},$
where σ² indicates the estimated variance and  indicates the estimated amplitude of the received signal.

[0027] The scaled first extrinsic information is obtained by dividing the first extrinsic information by

[0028] $\frac{\hat{A}{\sigma }^2}{2},$
where σ² indicates the estimated variance and  indicates the estimated amplitude of the received signal.

[0029] The scaled second extrinsic information is obtained by dividing the second extrinsic information by

[0030] $\frac{\hat{A}{\sigma }^2}{2},$
where σ² indicates the estimated variance and  indicates the estimated amplitude of the received signal.

[0031] The first decoder is a soft-on-soft-out decoder.

[0032] The second decoder is a soft-on-soft-out decoder.

[0033] The concatenation code is a turbo code.

[0034] Other objects, advantages, and novel features of the invention will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings.

[0044] FIG. 4 is a schematic diagram of a wireless transmission system in accordance with the invention. In FIG. 4, the wireless transmission system includes a turbo encoder 410, a modulator 420, a demodulator 430 and a turbo decoder 440, in which Gc indicates a wireless channel gains, n indicates an additive white Gaussian noise (AWGN) existed in the wireless channel, Ga indicates a system gain in a receiving system before the turbo decoder, and y indicates a receiving signal. The gain Gc is a constant in a code block and independent to each other in different code blocks. The AWGN noise n is Gaussian distribution with zero mean and variance σ_c². The received signal y can be expressed by the following equation:
y_i=G_aG_c√{square root over (E_s)}x_i+G_an=Ax_i+n′,
where the amplitude A=G_aG_c√{square root over (E_s)}, and the AWGN noise n′ is Gaussian distribution with zero mean and equivalent variance σ², where σ²=G_a²G_c². Therefore, the effective SNR is given by:

[0045] ${\mathrm{SNR}}_r=\frac{G_c^2{E}_s}{{\sigma }_c^2},$
and an effective noise variance σ² is defined by:

[0046] $\begin{array}{cc}{\sigma }^2\equiv \frac{1}{{\mathrm{SNR}}_r}=\frac{{\sigma }_c^2}{G_c^2{E}_s}.& \left(1\right)\end{array}$

[0047] Further, an input log-likelihood ratio (LLR) can be expressed by the following equation:

[0048] ${\Lambda }_{in,i}=\log \frac{P\left(y_i|x_i=+1\right)}{P\left(y_i|x_i=-1\right)}=\frac{2G_c\sqrt{E_s}}{G_a{\sigma }_c^2}{r}_i=\frac{2A}{{\sigma }^2}{r}_i.$

[0049] FIG. 5 is a block diagram of a decoding system 440 for real-time wireless channel estimation in accordance with the invention. The decoding system 440 decodes a received signal with a noise. The received signal y is coded by a concatenation code and has K message bits. The K message bits are grouped into systematic bits μ, first parity bits p and second parity bits q. The concatenation code is a turbo code. The LLRs Λ_in of the received signal y includes the LLRs of the system bits μ, first parity bits p and second parity bits q. Namely, we have Λ_in={Λ_μ, Λ_p, Λ_q}, where Λ_μ denotes the LLRs of the systematic bits μ, Λ_p denotes the LLRs of the first parity bits p, and Λ_q denotes the LLRs of the second parity bits q.

[0050] The system 440 includes a first decoder 510, an interleaver 520, a second decoder 530, a deinterleaver 540, a channel estimator 550, a controller 560 and a memory 570.

[0051] At the beginning of the decoding, the system 440 initializes an index to be 1 and the Prior₁⁽ⁱ⁾ to be 0, and sets s as a predetermined number of decoding.

[0052] The first decoder 510 is a soft-input-soft-output (SISO) decoder to receive a systematic bit log-likelihood ratio (LLR) Λ_μ, a first parity bit LLR Λ_p and a first prior information Prior₁⁽ⁱ⁾, and accordingly performs a Max-Log-MAP decoding. If the Max-Log-MAP decoding is performed, the first decoder 510 produces a first extrinsic information E₁⁽¹⁾ and a first output LLR Λ_out,1⁽¹⁾.

[0053] The interleaver 520 is connected to the first decoder 510 in order to receive the first extrinsic information (E₁⁽¹⁾) and the systematic bit LLR Λ_μ for re-arrangement to thereby produce a second prior information Prior₂⁽¹⁾ and an interleaved systematic bit LLR Λ_μ^π.

[0054] The second decoder 530 is a soft-input-soft-output decoder connected to the interleaver 520 in order to receive the interleaved systematic bit LLR Λ_μ^π, a second parity bit LLR Λ_q and the second prior information Prior₂⁽¹⁾, and accordingly performs the Max-Log-MAP decoding. If the Max-Log-MAP decoding is performed, the second decoder 530 produces a second extrinsic information E₂⁽¹⁾ and a second output LLR Λ_out,2⁽¹⁾.

[0055] The deinterleaver 540 is connected to the second decoder 530 in order to receive the second extrinsic information (E₂⁽¹⁾) for re-arrangement to thereby produce the first prior information Prior₁⁽²⁾. At this moment, the system 440 completes the first iteration of decoding and increases the index i with 1 to perform the second iteration of decoding.

[0056] The first stage decoding is performed with s iterations of iterative decoding. When the index i is increased to be equal to the predetermined decoding number s, the system 440 temporarily suspends the decoding and sends the LLRs (Λ_out,1^(s), and Λ_out,2^(s)) to the channel estimator 550 for channel estimation (Â and {circumflex over (σ)}).

[0057] The channel estimator 550 is connected to the first and the second decoders 510 and 530 in order to perform a channel estimation in accordance with the received signal, the first output LLR and the second output LLR to thereby produce estimated amplitude  and variance {circumflex over (σ)} of the received signal that are subsequently output to the first decode 510 and the second decoder 530.

[0058] The controller 560 is connected to the channel estimator 550, the first decoder 510 and the second decoder 530 in order to control the first and the second decoders 510 and 530 to perform a certain amount of decoding on the received signal by a Max-Log-MAP decoding procedure. Accordingly, the first extrinsic information and the first output LLR produced by the first decoder 510 are denoted by E₁^(s) and Λ_out,1^(s) respectively. The second extrinsic information and the second output LLR produced by the second decoder 530 are denoted by E₂^(s) and Λ_out,2^(s) respectively.

[0059] The controller 560 then controls the first decoder 510 and the second decoder 530 to use a Log-MAP decoding procedure in decoding. In this case, the controller 560 is based on a message length and a code rate to predetermine when the decoding procedure is changed from the Max-Log-MAP decoding procedure to the Log-MAP decoding procedure. Accordingly, the first extrinsic information and the first output LLR produced by the first decoder 510 are denoted by E₁⁽ⁱ⁾ and Λ_out,1⁽ⁱ⁾ respectively. The second extrinsic information and the second output LLR produced by the second decoder 530 are denoted by E₂⁽ⁱ⁾ and Λ_out,2⁽ⁱ⁾ respectively.

[0060] The memory 570 connected to the controller 560 stores a specific table. The controller 560 is based on the code rate of the concatenation code and the K value to look up the table and obtain the certain amount.

[0061] When the first decoder 510 and the second decoder 530 perform the Max-Log-MAP decoding procedure, there is no need to adjust the received signal in accordance with the estimated amplitude and variance A and a because of no need to understand the channel conditions. When the first decoder 510 and the second decoder 530 perform the Log-MAP decoding procedure, there is a need to adjust the received signal in accordance with the estimated amplitude and variance  and {circumflex over (σ)} because understanding the channel conditions is necessary, thereby producing an adjusted received signal. When the first decoder 510 and the second decoder 530 depends on the adjusted received signal to perform the Log-MAP decoding procedure, the first extrinsic information E₁⁽ⁱ⁾ produced by the first decoder 510 and the second extrinsic information E₂⁽ⁱ⁾ produced by the second decoder 530 are also adjusted in accordance with the estimated amplitude and variance  and {circumflex over (σ)} to further produce a scaled first extrinsic information and a scaled second extrinsic information.

[0062] FIG. 6 is a flowchart of a decoding method for real-time wireless channel estimation in accordance with the invention. The decoding method decodes a code division multiple access (CDMA) signal with a noise. The received signal is coded by a concatenation code. The concatenation code is a turbo code. Step (A) receives the received signal y with K message bits.

[0063] Step (B) performs a certain amount of decoding on the received signal y by the Max-Log-MAP decoding procedure. The certain amount is obtained by looking up a specific table based on the code rate of the concatenation code and the K value.

[0064] Step (C) uses the temporary LLR of each message bit of the received signal to compute the hard decision of each message bit of the received signal. Since step (B) performs the certain amount of Max-Log-MAP decoding procedure, the temporary LLR of i-th message bit is denoted by Λ_i^(s). The hard decision û_i^(s) of the i-th message bit in step (C) can be expressed by
û_i^(s)=0, if Λ_i^(s)≧0; and
û_i^(s)=1, if Λ_i^(s)<0,
where Λ_i^(s) indicates the temporary LLR.

[0065] Step (D) depends on the K message bits and the hard decision û_i^(s) of each message bit to compute an ensemble average of the K message bits. The ensemble average of the K message bits is regarded as an estimated amplitude  of the received signal. In this case, the estimated amplitude  of the received signal in step (D) is expressed by:

[0066] $\begin{array}{cc}\hat{A}=\frac{\stackrel{K}{\sum _i}{\left(-1\right)}^{{\hat{u}}_i^{\left(s\right)}}\times y_{u,i}}{K},& \left(2\right)\end{array}$
where û_i^(s) indicates a hard decision of i-th message bit, and y_u,i indicates the i-th message bit.

[0067] In this embodiment, the entire K message bits are used in step (D). However, in other embodiments, the partial message bits can be used in step (D). For example, the partial message bits

[0068] $\frac{K}{2},\frac{K}{3},100,\dots$
are used to compute the estimated amplitude Â. The partial message bits can be any message bits or highest reliable values selected from the K message bits. Namely, if the partial message bits are used in step (D), the estimated amplitude  of the received signal is expressed by:

[0069] $\begin{array}{cc}\hat{A}=\frac{\sum _i^K{\left(-1\right)}^{{\hat{u}}_i^{\left(s\right)}}\times y_{u,i}}{K^{\prime }},& \left(2^{\prime }\right)\end{array}$
where K′ can be

[0070] $\frac{K}{2},\frac{K}{3},\dots .$

[0071] In step (E), the ensemble average of a first momentum is computed in accordance with an absolute of the K message bits. The ensemble average W of the first momentum is expressed by:

[0072] $\begin{array}{cc}W=\frac{\sum _i^Ky_{u,i}}{K},& \left(3\right)\end{array}$
where |y_u,i| is an absolute of i-th message bit. If the partial message bits are used in step (D), the ensemble average W of the first momentum is expressed by

[0073] $\begin{array}{cc}W=\frac{\sum _i^Ky_{u,i}}{K^{\prime }},& \left(3^{\prime }\right)\end{array}$

[0074] Step (F) depends on the ensemble average W of the first momentum and the estimated amplitude  of the received signal to compute a ratio a, and computes the estimated variance {circumflex over (σ)} of the noise in accordance with the ratio a. The ratio a in step (F) is expressed by:

[0075] $a=\frac{W}{\hat{A}},$
where W indicates the ensemble of the first momentum, and  indicates the estimated amplitude of the received signal. The estimated variance {circumflex over (σ)} is computed by a polynomial, for example,
{circumflex over (σ)}²=−1.8833×a²+8.6671×a−6.5398,  (4)
where a is the ratio. In this embodiment, a second-order polynomial is used. However, in other embodiments, a polynomial of different order or a polynomial with coefficients approximated to those in equation (4) can be used.

[0076] When the estimated variance {circumflex over (σ)} of the noise is obtained in step (F), the channel quality is determined accordingly. Namely, the invention can be applied in a wireless channel estimation. Therefore, the invention has a considerably reduced computational complexity in comparison with the U.S. Pat. No. 6,393,257.

[0077] Step (G) depends on the estimated variance {circumflex over (σ)} of the noise and the estimated amplitude  of the received signal to adjust the received signal to thereby obtain an adjusted received signal y′. Subsequently, the Log-MAP decoding procedure is used to decode the adjusted received signal.

[0078] In step (G), the adjusted received signal is obtained by dividing the received signal or the last adjusted received signal y′ by

[0079] $\frac{\hat{A}{\sigma }^2}{2},$
where σ² indicates the variance, and  indicates the amplitude of the received signal.

[0080] FIGS. 7-1 and 7-2 show tables of computed and simulated results of the estimated amplitude  in accordance with the invention. The values of weighted extrinsic field and weighted LLR field in FIGS. 7-1 and 7-2 are obtained by the Monte Carlo simulation. As shown in FIGS. 7-1 and 7-2, the simulated results of the estimated amplitude  are very close to a true value A=1. In addition, the ones that are selected from the simulated results and have the highest accuracy can be regarded as the certain amount at different conditions.

[0081] FIG. 8 is a schematic graph of a bit error rate (BER) and signal-to-noise ratio (SNR) comparison in accordance with the invention. As shown in FIG. 8, the simulated results are obtained under the gains Gc and Ga to have a change range from −6 dB to +6 dB for rate ⅓ turbo codes with K=300, 1600, 5000 respectively, and the certain amount s=0.5, 1.0, 1.5, 2.0 respectively.

[0082] FIG. 9 is a schematic graph of another result of bit error rate (BER) and signal-to-noise ratio (SNR) comparison data in accordance with the invention. In this case, the simulated results are obtained under the gains Gc and Ga to have a change range from −6 dB to +6 dB for rate ⅗ turbo codes with K=300, 1600, 5000 respectively, and the certain amount s=0.5, 1.0, 1.5, 2.0 respectively.

[0083] As shown in FIGS. 8 and 9, it is known that the certain amount s=1.0 is the best choice for code rates ⅓ and ⅗ and K=1600 and 5000 respectively. In addition, the performance is reduced to a value below 0.1 dB at different certain amount s, which is caused by the difference between the Max-Log-MAP decoding procedure and the Log-MAP decoding procedure, but not by the accuracy of the estimated amplitude  in the invention.

[0084] In view of forgoing, it is known that, in the invention, when the first and second decoders 510 and 530 perform the Log-MAP decoding procedure, the channel conditions  and {circumflex over (σ)} are known, and accordingly it is able to avoid that the Log-MAX decoding has a poor performance in comparison with the Max-Log-MAX decoding when the inaccurate channel estimation occurs.

[0085] Although the present invention has been explained in relation to its preferred embodiment, it is to be understood that many other possible modifications and variations can be made without departing from the spirit and scope of the invention as hereinafter claimed.