JP4191697B2 - Turbo receiving method and receiver thereof - Google Patents

Description

  The present invention relates to a turbo reception method that is applied to, for example, mobile communication and performs waveform equalization based on interference to perform equalization repeatedly using a turbo code technique, and a receiver thereof.

The problem of the mobile communication business is how to build a system that can own a large number of users with high quality on a limited frequency. As a means for solving such a problem, there is a multi-input multi-output (MIMO) system. In this system configuration, as shown in FIG. 30A, symbols c ₁ (i) to c _N (i) are transmitted from a plurality of transmitters S1 to SN at the same time and the same frequency. Received by a MIMO receiver having a plurality of antennas # 1 to #M, the MIMO receiver processes the received signal, and estimates the transmission symbols c ₁ (i) to c _N (i) of the respective transmitters S1 to _SN Then, c ₁ ^ (i) to c _N ^ (i) are separately output to the output terminals Out 1 to OutN.

To date, there has not been a sufficient study on a specific configuration method of a MIMO receiver in a MIMO system. When the configuration of the MIMO receiver in the MIMO system is performed based on the MLSE (maximum likelihood estimation) standard, the number of transmitters is N, and the number of multipaths in which the transmission radio waves of each transmitter reach the MIMO receiver is Q. Then, the calculation amount of the MIMO receiver becomes a digit of 2 ^{(Q-1) N} , and the calculation amount becomes enormous as the number of transmitters N and the number of multipaths Q increase. Further, when receiving single user information transmitted as a plurality of parallel signals, separating the parallel signals requires a large amount of calculation as the number of multipaths increases. Therefore, the present invention proposes a turbo receiving method of a plurality of sequence signals with high computational efficiency. First, an existing single user (one transmitter) as a basis of the present invention, that is, a turbo receiver for one sequence transmission signal. Will be described.

Single-User Turbo Receiver FIG. 31 shows a configuration example of a transmitter and a receiver in this case. In the transmitter 10, the information sequence c (i) is encoded by the encoder 11, and the encoded output is interleaved (reordered) by the interleaver 12, and then the carrier signal is modulated by the modulator 13. A modulated output is transmitted. This transmission signal is received by the receiver 20 through a transmission path (multipath channels). In the receiver 20, the delay wave is equalized by a soft-soft-soft-output (SISO: Single-Input-Single-Output) equalizer 21. The input of the equalizer 21 is generally converted into a digital signal by converting the received signal into a baseband, and the baseband received signal is sampled at a frequency of 1 or more times the frequency of the symbol signal of the information sequence of the transmission signal. And is input to the equalizer 21 as a digital signal reception signal.

In the case of a single user, N = 1 in FIG. 30A, and the reception output at each reception antenna #m (m = 1, 2,..., M)
r _m (k) = Σ _{q = 0} ^Q−1 h _m (q) b (k−q) + v _m (k) (1)
It can be expressed as. m is an antenna index, h is a channel value (transmission path impulse response: transmission path characteristics), b (k−q + 1) is a transmission symbol of the user (transmitter 1), and v _m (k) is the inside of the receiver 20. The thermal noise. The outputs from all the antennas # 1 to #M are expressed as vectors of the equation (2), and the equation (3)
r (k) = [r ₁ (k) r ₂ (k)... r _M (k)] ^T (2)
= Σ _{q = 0} ^Q-1 H (q) b (k−q + 1) + v (k) (3)
Define here,
v (k) = [v ₁ (k) v ₂ (k)... v _M (k)] ^T (4)
H (q) = [h ₁ (q)... H _M (q)] ^T (5)
It is. [] ^T represents a transposed matrix. Next, the following vectors and matrices are defined in consideration of the number Q of multipaths (channels).

y (k) = [r ^T (k + Q-1) r ^T (k + Q-2)... r ^T (k)] ^T
(6)
≡H · b (k) + n (k) (7)
here,

However,
b (k−q) = [b (k + Q−1) b (k + Q−2)... b (k−Q + 1)] ^T
(9)
^{n (k) = [v T} (k + Q-1) v T (k + Q-2) ... v T (k)] T
(Ten)
And
The r (k) defined above is input to the equalizer 21. The SISO equalizer 21 is a linear equalizer, and each encoded bit {b (i)} is +1 as an equalization output. A log likelihood ratio [Lambda] ₁ (LLR: Log-Likelihood Ratio) of a probability and a probability of -1 is derived.

It is. Here, λ ₁ [b (k)] is external information sent to the subsequent decoder 24, and λ ₂ ^p [b (k)] is a priori information given to the equalizer 21. The log likelihood ratio Λ ₁ [b (k)] is subtracted from the prior information λ ₂ [b (k)] by the subtracter 22 and further supplied to the SISO channel decoder 24 via the deinterleaver 23. The decoder 24 has a log likelihood ratio Λ ₂ ,

Is calculated. Here, λ ₂ [b (i)] is external information given to the equalizer 21 as λ ₂ ^p [b (k)] in the iteration, and λ ₁ [b (k)] is sent to the decoder 24. It is given as prior information λ ₁ ^p [b (i)]. Λ ₂ [b (i)] is subtracted from λ ₁ [b (i)] by the subtractor 25 and supplied to the equalizer 21 and the subtractor 22 via the interleaver 26. In this way, iterative equalization and decoding are performed to improve the error rate.
Next, calculation of the linear filter characteristic applied to the reception vector y (k) will be described as details of the equalizer 21 in the previous stage. Soft decision symbol estimate b ′ (k) = tan h [λ ₂ ^p [b (k)] / 2] using prior information λ ₂ ^p [b (k)] of equalizer 21 (15)
Is calculated. Then, using this estimated value and the channel matrix H, an interference component, that is, a replica H · b ′ (k) of the interference component is reproduced and subtracted from the received signal. That is, y ′ (k) ≡y (k) −H · b ′ (k) (16)
= H. (B (k) -b '(k)) + n (k) (17)
here,
b '(k) = [b' (k + Q-1) ... 0 ... b '(k-Q + 1)] ^T (18)
Calculate Since the replica H · b ′ (k) of the interference component is not necessarily an exact replica, the interference component cannot be completely removed by the equation (16). Therefore, a linear filter coefficient w (k) that eliminates the remaining interference component is obtained according to the following MMSE (minimum mean square error) standard.

w (k) = arg min ‖w H (k) · y '(k) -b (k) || ² (19)
^H represents a conjugate transpose, and ‖ ノ ル represents a norm.
Find w (k) that minimizes Equation (19).
The following derivation of w (k) is described in Non-Patent Document 1. The main achievement of this method is a significant reduction in computational complexity. Calculated amount of conventional MLSE turbo Whereas was proportional to 2 ^Q-1 of the order, this approach is suppressed on the order of Q ^3. W ^H (k) · y
′ (K) is the output of the equalizer 21, from which λ ₁ [b (k)] is calculated and supplied to the decoder 24 via the deinterleaver 23 to perform a decoding operation.

  In order to perform equalization processing in the equalizer 21, it is necessary to estimate the channel value (transmission path impulse response) h in the equation (1). This channel value estimation is hereinafter referred to as channel estimation. Channel estimation is performed using a received signal of a known training sequence such as a unique word sent to the head of one frame and a stored training sequence. If the accuracy of channel estimation is poor, the equalization process in the equalizer 21 is not performed correctly. In order to increase the accuracy of channel estimation, the proportion of the training sequence in one frame may be increased, but if so, the transmission efficiency for the original data is reduced. Therefore, it is desirable to reduce the proportion of training sequences in one frame and improve channel estimation accuracy.

This is not limited to receivers for multi-sequence transmission signals including MIMO, but in receivers that use a rake receiver or an adaptive array antenna to improve the accuracy of decoding results by iterative decoding processing, There is a similar problem in channel estimation.
1Daryl Reynolds and Xiaodong Wang, “Low Complexity Turbo-Equalization for Diversity Channels” (http: /ee.tamu.edu/reynolds/)

  In conventional turbo reception, the channel estimation is performed only by the training signal. In this respect, channel estimation is not always sufficient, and there is a problem in that it is necessary to increase the number of repetitions for correct symbol estimation.

  According to the present invention, the probability of the decoded hard decision information symbol is determined from the value of the soft decision information symbol, and a hard decision information symbol having a certainty or more of a certain value is also used as a reference signal for the next channel estimation. .

  Since a reliable hard decision information symbol is also used as a reference signal for the next channel estimation, more accurate channel estimation can be performed, and the number of iterative decoding can be reduced accordingly.

1st invention (1)
FIG. 1 shows a configuration example of a MIMO system to which the present invention is applied.
Information sequences c ₁ (i)... C _N (i) are respectively encoded by encoders 11-1,..., 11-N in each of N transmitters S1. .., 12-N are supplied as modulation signals to modulators 13-1,..., 13-N through interleavers 12-1,..., 12-N, and carrier signals are modulated by these modulation signals to generate signals b ₁ (k) _,. It is transmitted as (k). That is, the transmission signals b ₁ (k),..., B _N (k) from the transmitters S1,.

The received signal r (k) received by the multi-output receiver through the transmission path (channel) is input to the multi-output equalizer 31, and the signal received by the receiver is converted into a baseband signal. For example, the signal is sampled at half the symbol period, converted into a digital signal, and input to the equalizer 31 as the digital signal. Further, the number of digital signals is an integer M equal to or greater than 1. For example, reception signals from M antennas are used as reception signals of M digital signals. N log likelihood ratios Λ ₁ [b ₁ (k)],..., Λ ₁ [b _N (k)] are output from the equalizer 31. _{Λ 1 [b 1 (k)} ], ..., Λ 1 [b N (k)] , respectively priori information _{λ 1 [b 1 (k)} ], ..., λ 1 [b N (k)] is the subtractor 22 -1, ..., 22
-N, respectively, and are respectively input to soft input / soft output (SISO) decoders (channel decoders) 24-1, ..., 24-N through the deinterleavers 23-1, ..., 23-N and decoded. Decoding information sequences c ′ ₁ (i),..., C ′ _N (i) are output from the decoders 24-1,..., 24-N, and log likelihood ratios Λ ₂ [b ₁ (i)],. , Λ ₂ [b _N (i)] are respectively output. _{Λ 2 [b 1 (i)} ], ..., Λ 2 [b N (i)] is a subtractor 25-1, ..., ₁ λ by _{25-N [b 1 (i} )], ..., λ 1 [b _N (i)] are subtracted, and further output as λ ₂ [b ₁ (k)],..., Λ ₂ [b _N (k)] through the interleavers 26-1,. , 22-N are supplied to the equalizer 31 and the subtracters 22-1,.

Received signal r _m from the multi-user (s transmitters) (k) (m = 1 , ..., M) as an input of the equalizer 31,
r _m (k) = Σ _{q = 0} ^Q−1 Σ _{n = 1} ^N h _mn (q) · b _n (k−q) + v _m (k) (20)
And multiple users. q = 0,..., Q−1, Q are the number of multipaths of each transmission radio wave, and the vector y (k) is defined in the same procedure as in the case of a single user.
y (k) ≡ [r ^T (k + Q−1) r ^T (k + Q−2)... r ^T (k)] ^T (21)
= H · B (k) + n (k) (22)
Here, r (k) = [r ₁ (k)... R _M (k)] ^T

ただし、

However,

B (k) = [b ^T (k + Q−1)... B ^T (k)... B ^T (k−Q + 1)] ^T (25)
b (k + q) = [b ₁ (k + q) b ₂ (k + q)... b _N (k + q)] ^T
q = Q-1, Q-2, ..., -Q + 1 (26)
It becomes.
Now assume that in the interference cancellation step, the signal from the nth user (transmitter) is now desired. In this example, using the soft-decision symbol estimation values of signals from all users (transmitters) and the channel matrix (transmission path impulse response value matrix) H, interference by signals of users other than the n-th user and the n-th user And the interference replica H · B ′ (k) is reproduced and the interference replica is subtracted from y (k) as follows to obtain the difference vector y ′ (k ) Is generated.

y ′ (k) ≡y (k) −H · B ′ (k) (27)
= H. (B (k) -B '(k)) + n (k) (28)
here,
B ′ (k) = [b ′ ^T (k + Q−1)... B ′ ^T (k)... B ′ ^T (k−Q +)] ^T (29)
And
b ′ (k + q) = [b ′ ₁ (k + q) b ′ ₂ (k + q)... b ′ _N (k + q)] ^T : q = Q−1,..., −Q + 1, q ≠ 0 (30)
b ′ (k) = [b ′ ₁ (k)... 0 b ′ _N (k)] ^T : q = 0 (31)
0 in the element of b '(k) is the nth.
b ′ _n (k) is a soft-decision transmission symbol estimation value obtained by calculating b ′ _n (k) = tan h [λ ₂ [b _n (k)] / 2] similarly to the equation (15). . The vector B ′ (k) is an interference symbol replica vector.

Next, for the nth user for canceling the remainder of the interference component, that is, the interference residue based on the imperfection of the interference component replica H · B ′ (k) and the interference component produced by the nth signal itself. The filter coefficient w _n (k) is determined by MMSE (minimum mean square error) norm, w _n (k) that minimizes the following equation (32).
_{w n (k) = arg min} ‖w n H (k) · y '(k) -b n (k) || ² (32)
The following operations are the same as those for a single user. That is, w _n ^H (k) · y ′ (k) is calculated using the obtained w _n (k), and the calculation result is set as λ ₁ [b _n (i)] via the deinterleaver 23-n. A decoding operation is performed by inputting to the decoder 24-n.

To seek a filter (linear equalization) process on the received signal r _m from the user 1 in the above manner until N. As a result, the number of outputs of the equalizer 31 is N, and these outputs are decoded by the respective decoders 24-1,. The above is the extension of the single-user turbo receiver to multi-user (MIMO).
From the above description, an example of the functional configuration of the multi-output equalizer 31 is as shown in FIG. A reception vector y (k) is generated from the M reception signals r _m (k) by the reception vector generation unit 311 and is supplied to the equalization units 312-1 to 312-N for each user.
The channel matrix H calculated in the channel estimator 28 is also supplied to the equalization units 312-1 to 312-N. Prior information λ ₂ [b _n (k)] from each channel decoder 24-n is input to soft decision symbol estimation section 313, and soft decision transmission symbol estimate b ′ _n (k) = tanh [λ ₂ [ b _n (k)] / 2] is calculated. The functional configuration and processing in the equalization units 312-1 to 312-N are the same and will be described as a representative of the equalization unit 312-1.

Further, the estimated values b ′ ₁ (k) to b ′ _N (k) of the soft decision transmission symbols are supplied to the interference replica vector generation unit 314-1, and the interference replica vector B ′ _{1 is} calculated according to the equations (29) to (31). (K) is generated, the vector B ′ ₁ (k) is filtered by the channel matrix H in the filter processing unit 315-1, and the resulting interference replica component H · B ′ ₁ (k) is converted into the difference calculation unit. At 316-1, a difference vector y ′ ₁ (k) is generated by subtracting from the received vector y ₁ (k).

At least the channel matrix H or a reference signal as described later is input to the filter coefficient estimation unit 317-1, and a filter coefficient w ₁ (k) for removing the remainder of the interference component is obtained. In this example, the channel matrix H from the channel estimator 28 and the noise component covariance σ ² and the soft decision transmission symbols b ′ ₁ (k) to b ′ _N (k) from the soft decision symbol estimator 313-1 are: A filter coefficient w ₁ (k) that is input to the filter coefficient estimation unit 317-1 and minimizes the expression (32) is obtained using the minimum mean square error criterion. Specific processing for obtaining the filter coefficient w ₁ (k) will be described later. The differential vector y ′ ₁ (k) is processed by the filter coefficient w ₁ (k) in the adaptive filter processing unit 318-1, and Λ ₁ [b ₁ (k) is used as the equalized output of the received signal with respect to the transmission signal from the user 1. ] Is output.

FIG. 3 shows the processing procedure of the multi-input multi-output turbo receiving method of the embodiment of the present invention described above. In step S1, a channel value h _mn (q) and a noise component covariance σ ² are calculated from the received signal r (k) and each training signal b _n (k). In step S2, the channel value h _mn (q) is calculated. The channel matrix H is calculated, and the soft decision transmission symbol estimated value b ′ _n (k) = tanh (λ ₂ ) from each prior information λ ₂ [b _n (k)] obtained in the previous process in the turbo reception process in step S3. [B _n (k)] / 2) is calculated.

It generates a reception vector y (k) from the received signal r (k) in step S6, the interference replica by equation (29) to (31) using the respective soft decision transmitted symbol estimate b _'n (k) in step S3 A vector B ′ _n (k) is generated (step S4), and an interference component replica H · B ′ _n (k) for the received signal from the nth transmitter is calculated in step S5. In step S7, the interference component replica H · B ′ _n (k) is subtracted from the received vector y (k) to obtain a difference vector y ′ _n (k). In step S8, the residual interference in the received signal from the nth transmitter is determined by the channel matrix H, the soft decision transmission symbols b ′ ₁ (k) to b ′ _N (k), and the noise component covariance σ ^2. The filter coefficient w _n (k) for removal is obtained by the minimum mean square error criterion that minimizes the equation (32).

In step S9, the difference vector y ′ _n (k) is filtered by the filter coefficient w _n (k) to obtain the log likelihood ratio Λ ₁ [b _n (k)]. In step S10, prior information λ ₂ [b _n (k)] is subtracted from Λ ₁ [b _n (k)], deinterleaving is performed, and decoding is performed to obtain a log likelihood ratio Λ ₂ [b _n (k). )] Is output. The processes in steps S4 to S10 are performed simultaneously or sequentially for n = 1 to N. Thereafter, in step S11, it is checked whether the number of decoding, that is, the number of turbo reception processes, has reached a predetermined number. If not, in step S12, the log likelihood ratio Λ ₂ [b _n (k)] is used to obtain external information λ. ₁ [b _n (k)] is subtracted, and the result is interleaved to obtain prior information λ ₂ [b _n (k)], and the process returns to step S3. If the decoding is a predetermined number of times in step S11, the decoding result at that time is output in step S13.

Next, the channel estimation unit 28 will be described. Each received signal r _m (k) can be expressed by the following equation.
r _m (k) = Σ _{q = 0} ^Q−1 Σ _{n = 1} ^N h _mn (q) · b _n (k−q) + v _m (k) (33)
The channel estimation unit 28 obtains the value of h _mn (q) of the channel value (transmission path impulse response) and the average power (≡σ ² ) of the noise v _m (k) in Expression (33). As shown in FIG. 4A, the normal transmission side inserts a unique word (training signal) known by the receiver at the beginning of each transmission frame, and the receiver uses the unique word (known signal) as a training sequence to perform RLS (recursive minimum). The channel value h _mn (q) is estimated using a square method. Each channel decoder 24-1, ..., from 24-N, the log-likelihood ratio _{Λ 2 [b 1 (i)} ], ..., for each of _{Λ 2 [b N (i)} ], if positive +1 -1 are output as decoded code signals (transmission coded symbol hard decision values) b ₁ ^ (i),..., B _N ^ (i), respectively, and b ₁ ^ (i),. b _N ^ (i) is repeatedly input to the channel estimator 28 through the interleavers 27-1,. The channel estimator 28 receives the received signal r (k) and the unique word from the unique word storage unit 29 as a reference signal. Based on these input signals, the channel estimator 28 estimates each value of h _mn (q) and σ ² in equation (33) by the least square method. This estimation can be performed by a method similar to the estimation of the impulse response when the impulse response of the transmission path is estimated and the received signal is adaptively equalized by the adaptive filter.

Using a training sequence in this way is a commonly used technique, but to increase the net transmission rate, it is necessary to reduce the proportion of unique words in one frame, which increases the error in channel estimation. To do. The error degrades the above-described repetitive equalization characteristics. Therefore, it is recommended that the channel value is repeatedly estimated as follows.
The concept is shown in FIG. 4B. This is to repeatedly estimate the channel value at each stage of the repeated equalization processing of the same received signal, that is, the repeated processing of the turbo reception processing. That is, at the first time, for the information symbol sequence after the unique word, the channel value is estimated using only the unique word as a reference signal, and the received signal is equalized using the estimated channel value to estimate the transmission symbol. However, before the second and subsequent equalization processes, channel estimation is performed using the unique word as a reference signal, and the symbol estimation value (hard decision value) obtained in the previous decoding process is also used as a reference signal. Channel estimation in the entire frame. In this case, not all the hard decision values are used, but only the hard decision values determined to be reliable may be used as reference signals. A hard decision is made by using the log-likelihood ratio Λ ₂ [b _n (i)] from the decoder 24-n, +1 if this is positive, and -1 if negative. At that time, it can be said that the hard decision value is more likely as the absolute value of the log likelihood ratio Λ ₂ [b _n (i)] is larger. For example, 1 when log likelihood 5 is determined to be 1 is more likely than 1 when log likelihood 0.3 is determined to be 1. Therefore, a method for selecting a probable hard decision value b _n (i) using a threshold value and repeatedly performing channel estimation using the hard decision value b _n (i) will be described below.

First, using the log likelihood ratio Λ ₂ [b _n (i)] from the decoder 24-n, the soft decision value b ′ _n (i) of the symbol is
b ′ _n (i) = tan h [Λ ₂ [b _n (i)] / 2]
Asking. This operation is to normalize the log likelihood value to 1 so that the absolute value does not exceed 1. Next, a threshold value (between 0 and 1) is prepared in advance, and the hard decision value b ^ for the soft decision value b ′ _n (i) whose absolute value is larger than the threshold value. _n (i) is stored and used repeatedly for channel estimation. For example, only by setting the threshold to 0.9 soft decision value b _'n (i) the absolute value of 0.9 or more hard decision values b ^ _n of (i) is selected. Since the threshold value is as high as 0.9, the accuracy of the selected hard decision value b ^ _n (i) is considered to be high. Accordingly, the number of symbols to be selected decreases, so that it is considered that the repeated channel estimation accuracy is lowered. In other words, it is necessary to select an optimum threshold value between 0 and 1. As a supplement, if the threshold value is set to 1, there is no hard decision value b ^ _n (i) to be selected, so that repeated channel estimation is not performed. Therefore, as will be described later, the threshold value is set to about 0.2 to 0.8.

Therefore, the symbol values determined to be probable by the threshold values in the transmission symbol estimation values (hard decision values) b ₁ ^ (i),..., B _N ^ (i) for the first information symbol sequence are interleaver 27 −. 1,..., 27-N are stored as previous transmission symbol estimation values in the previous symbol storage unit 32, and in the second iterative equalization decoding process of the reception signal r (k) (reception signal r (k ) Is stored in the storage unit), first, channel estimation is performed using a unique word, and further, an estimated transmission symbol hard decision estimated value b ^ ₁ (i) from the previous symbol storage unit 32 for the information symbol series. , ..., are input to b ^ _N (i) probable judged channel estimator 28 reads out the symbol values in, performs channel estimation, i.e. performs channel estimation for the entire frame, the With value h _mn (q) and sigma ^2, the equalization for the received signal r (k), decoding (transmission symbol estimation) performed. At this time, the stored content of the previous symbol storage unit 32 is updated with a symbol value determined to be probable by the estimated threshold value in the transmission symbol. In the same manner, channel estimation is performed for the entire frame by estimation using a unique word and estimation using a probable decision in the previous estimated transmission symbol in repetition of equalization and decoding. Do. The estimation channel is used for equalization and decoding (transmission symbol estimation), and the previous symbol storage unit 32 is updated. The previous symbol storage 32 stores the symbol values determined to be probable by the threshold values in the transmission symbol hard decision values b ₁ (i),..., B _N (i) from the decoder. When the stored symbol value in the storage unit 32 is directly updated and the stored symbol value of the previous symbol storage unit 32 is used, it may be input to the channel estimator 28 through the interleavers 27-1,.

By doing so, channel estimation errors are reduced by repetition, symbol estimation accuracy is improved, and the problem of characteristic deterioration due to channel estimation errors in turbo equalization can be improved.
When channel estimation is performed in an information symbol sequence using such a reliable symbol hard decision value, the functional configuration shown in FIG. 5 is added to each decoder 24-n. The log likelihood ratio Λ ₂ [b _n (i)] is input to the soft decision value estimator 241, b ′ _n (i) = tanh (Λ ₂ [b _n (i)]) is calculated, and the transmission symbol soft Determination value b ′ _n (i) is estimated, and this value b ′ _n (i) is compared with threshold value Th from threshold setting unit 243 by comparison unit 242, and b ′ _n (i) is _{equal to} or greater than Th. Is 1 and 0 is output when smaller than Th. On the other hand, the log likelihood ratio Λ ₂ [b _n (i)] is input to the hard decision unit 244. If Λ ₂ [b _n (i)] is positive, the symbol hard decision value b ^ is +1. _n (i) is output, and this symbol hard decision value b ^ _n (i) is output when the corresponding symbol soft decision value is greater than or equal to the threshold value, and the gate 245 is opened, and the interleaver in FIG. 27-n is supplied to the previous symbol storage unit 32, and the stored symbol is updated.

Further, a channel estimation procedure using a probable symbol hard decision value is as shown in FIG. First, in step S1, channel estimation based on the received signal r (k) and the unique word is performed. In step S2, whether or not the decoding process is the first time is checked. If it is the first time, the estimated channel value h _mn is determined in step S3. (Q) is used to perform equalization and decoding processing, that is, steps S3 to S10 in FIG.
In step S4, a transmission symbol hard decision process is performed on the log likelihood ratio Λ ₂ [b _n (i)] to obtain a hard decision value b ^ _n (i). In step S5, the log likelihood ratio Λ ₂ [b _n (i)], b ′ _n (i) = tanh (Λ ₂ [b _n (i)] / 2) is calculated to estimate the transmission symbol soft decision value b ′ _n (i). A probable symbol hard decision value b ^ _n (i) is determined depending on whether or not the symbol soft decision value b ′ _n (i) is _{equal to} or greater than the threshold Th in step S6, and the probable symbol is determined in step S7. The content stored in the previous symbol storage unit 32 is updated with the hard decision value. Next, it is checked in step S8 whether the number of times of decoding is a predetermined value. If not, the process returns to step S1. More precisely, the process returns to step S1 in FIG. 3 through step S12 in FIG.

If it is determined in step S2 that the decoding process is not performed once, the previous stored symbol, that is, a probable hard decision symbol is read from the previous symbol storage unit 32 in step S9, and this is the information symbol sequence of the received signal r (k). The channel estimation is performed using and and the process proceeds to step S3.
In the above, in the second and subsequent processing, channel estimation is performed from the initial state using the unique word as a reference signal. However, only a probable hard decision symbol may be used as the reference signal after the second processing. In this case, as indicated by a broken line in FIG. 6, it is checked whether the process is the first process in step S1 ′. If it is the first process, the unique word is used as a reference signal in step S2 ′ and the received signal. The channel value is estimated. In step S3 ′, the estimated channel value and each parameter value used for the estimation are stored in the storage unit. Then, the process proceeds to equalization and decoding processing in step S3.

If it is not the first time in step S1 ′, prior to the channel estimation process, the channel estimation value and various processing parameters previously stored in step S4 ′ are set, and the process proceeds to step S9.
The solution of equation (32) is
w _n (k) = (HG (k) H ^H + σ ² I) ⁻¹ · h (34)
I is the unit matrix, σ ² is the internal noise power (noise component covariance) of the receiver, σ ² I is the noise component covariance matrix, and G (k) corresponds to the channel estimation square error.

G (k) = E [(B (k) -B '(k)). (B (k) -B' (k)) ^H ]
= Diag [D (k + Q-1), ..., D (k), ..., D (k-Q + 1)]
(35)
E [] represents an average, and diag represents a diagonal matrix (elements other than diagonal elements are zero).
Also, D (k + q) = diag [1-b ′ ² ₁ (K + q), ..., 1-b '2 n (k + b), ..., 1-b' 2 N (k + q)] (36)
q = Q-1, Q-2,..., -Q + 1, q ≠ 0
When q = 0, D (k) = diag [1-b ′ ² ₁ (k),..., 1,. ² _N (k)] (37)
1 in the vector D (k) is the n-th element (the transmission signal of the n-th user is a desired signal).

That is, h is composed of all the elements in the (Q−1) · N + n column of H in Expression (23). As shown in FIG. 2, in the filter coefficient estimation unit 317-1 of the multi-output equalizer 31, the channel matrix H and noise power σ ² estimated by the channel estimator 28 and the soft decision symbol generation unit 313-1 Soft decision transmission symbols b ′ ₁ (k) to b ′ _N (k) are input, and the residual interference cancellation filter coefficient w _n (k) is calculated by the equations (34) to (38).
Equation (34) performs an inverse matrix operation, but this operation can reduce the amount of operation by using the inverse matrix theorem (Matrix Inversion Lemma). In other words, when each b ′ ² part of the equations (36) and (37) is approximated to 1,
D (k + q) = diag [0,..., 0] = 0 (q ≠ 0) (39)
D (k) = diag [0, ..., 1, ..., 0] (40)
That is, only the element of n rows and n columns in the element of D (k) is 1, and all other elements are 0. When the error matrix G (k) of the equation (35) determined by these equations (39) and (40) is substituted into the equation (34),
w _n (k) = (h · h ^H + σ ² I) ⁻¹ · h (41)
It becomes. h is defined by equation (38).

By this approximation, w _n (k) does not depend on k, so that an inverse matrix calculation for each discrete time k is not required, and the amount of calculation is reduced.
The inverse matrix lemma is applied to this equation (41). The inverse matrix theorem is that A and B are (M, M) square matrices, C is an (M, N) matrix, D is a (N, N) square matrix, and A = B ⁻¹ + CD ⁻¹ C ^When represented by ^H , the inverse matrix of A is: A ⁻¹ = B−BC (D + C ^H BBC) ⁻¹ C ^H B (42)
Given in. Applying this theorem to the inverse matrix operation part in equation (41),
h (k) · h (k) ^H + σ ² I = B ⁻¹ + CD ⁻¹ C ^H
h (k) · h (k) ^H = CD ⁻¹ C ^H , σ ² I = B ⁻¹ , h (k) = C
I = D ⁻¹ , h (k) ^H = C ^H
If the equation (42) is calculated using this, the inverse matrix operation in the equation (41) is obtained. Note that the inverse matrix operation (D + C ^H BBC) ⁻¹ is also included in the equation (42), but since this inverse matrix is a scalar, it can be easily calculated.

So in this case,
w _n (k) = 1 / (σ ² + h ^H · h) h (41-1)
It becomes. 1 / () on the right side of this equation is a scalar, that is, a constant number, so it may be 1. Therefore, since w _n (k) = h can be set, (K) is determined. 2 only needs to be input from the channel estimator 28 as indicated by the equation (38) in the channel matrix H, as indicated by a broken line.

Note that the approximation by the equations (39) and (40) is not limited to the case where the inverse matrix theorem is used, and the amount of calculation of the equation (34) can be reduced by this approximation. In particular, if this approximation is performed and the inverse matrix auxiliary theorem is used, the amount of calculation can be further reduced. In this case, if the covariance matrix of the noise component is σ ² I, w as shown in equation (41-1) It can be approximated by _n (k) = h, becomes irrelevant to the covariance matrix, and further simplifies the calculation.
Second invention (reflecting error correction)
In the equalization process of subtracting H · B ′ (k) from the received vector y (k) shown in Expression (27), the transmission symbol soft decision _values of signals other than the detected signal b _n (k) are error correction decoded. Although the result is reflected, the error correction decoding result regarding the detected signal b _n (k) is not reflected. Therefore, it is preferable to perform the following processing.

B '(k) in the equation (29), that is, the equation (31) is changed to the following equation.
b ′ (k) = [b ′ ₁ (k) b ′ ₂ (k)...
b ′ _n−1 (k) −f (b ′ _n (k)) b ′ _{n + 1} (k)...
b ′ _N (k)] (43)
However, f (b ′ _n (k)) is an arbitrary function having b ′ _n (k) as an input. By doing so, the error correction decoding result is reflected also on the detected signal b _n (k). Is possible. That is, by adding an appropriate value according to (b ′ _n (k) without setting b ′ _n (k) = 0, for example, a signal to be detected that is buried in noise or an interference signal is emphasized. As a result, b _n (k) can be detected correctly.

'For _{(n (k), b f} b)' the absolute value of the code of _n (k) is b 'relate to the hard decision result of the symbol corresponding to the _n (k), also b' _n (k) is The larger the value, the higher the reliability of the hard decision symbol corresponding to b ′ _n (k).
When b ′ _n (k) = 0, that is, when the reliability of the hard decision symbol is 0, the value of this function f is also 0. That is, f (0) = 0 (44)
It is. Further, if the value of b ′ _n (k) is large, the value of the function f is also large. That is, d {f (b ′ _n (k))} / d {b ′ _n (k)} ≧ 0 (45)
It is. Examples of such f (b ′ _n (k)) are:
f (b ′ _n (k)) = α × b ′ _n (k) (46)
f (b ′ _n (k)) = α × b ′ _n (k) ² (47)
Is mentioned. For example, if α is a constant using equation (46), equation (43) can be easily realized. Here, α is 0 <α <0.6. On the other hand, if α is larger than 0.6, the BER (error rate) characteristic deteriorates, and a correct decoding result cannot be obtained. It is also conceivable to vary α according to the reliability of the decoding result. For example, α is set every time the decoding process is repeated. In this case, the reliability of the decoding result usually increases as the number of repetitions of the decoding process increases. Therefore, the value of α may be increased according to the number of repetitions of the decoding process. Alternatively, the reliability of the entire decoded frame may be determined every time the decoding process is repeated, and the value of α may be determined based on the determination. As a method of determining the reliability of the decoded frame, for example, a method of comparing the decoding result with a decoding result at the previous iterative decoding and counting the number of hard decision symbols changed from the previous decoding can be considered. That is, when the number of changed hard decision symbols is large, it is determined that the reliability is low, and when the number of changed hard decision symbols is small, it is determined that the reliability is high.

Further, in accordance with such a change of b ′ _n (k), equation (35) used when obtaining the coefficient w _n (k) of the MMSE (minimum mean square error) filter can be changed as follows. desirable.
G (k) = E [(B (k) −B ′ (k)) · (B (k) −B ′ (k)) ^H ]
= Diag [D (k + Q-1), ..., D (k), ..., D (k-
Q + 1)]
Here, from Equation (29) and Equation (31)

B ′ (k) = [b ′ (k + Q−1)... B ′ (k)... B ′ (k−Q + 1)] ^T
b ′ (k + q) = [b ′ ₁ (k + q) b ′ ₂ (k + q)... b ′ _N (k + q)] ^T q = Q−1,..., −Q + 1, q ≠ 0 and b ′ (k) = [ b ′ ₁ (k)... -f (b ′ _n (k))... b ′ _N (k)] ^T
When q = 0, −f (b ′ _n (k)) is the n-th element of b ′ (k).

And The element of n rows and n columns of D (k) is E [(b _n (k) + f (b ′ _n (k))) · (b _n (k) + f (b ′ _n (k)))] ^*
[] ^* Represents a complex conjugate. This equation becomes the following equation in the case of BPSK modulation.
E [b _n (k) ² + 2b _n (k) f (b ′ _n (k)) + f (b ′ _n (k)) ² ]
= E [b _n ² (k)] + 2E [b _n (k) f (b ′ _n (k)] + E [f (b ′ _n (k) ² ]]
The average value of this first term is 1. Further, when b _n (k) is approximated by b ′ (k), Expression (37) is as follows.

D (k) = diag [1-b ′ ² ₁ (k) 1-b ′ ² ₂ (k)...
1-b ′ ² _n−1 (k) 1 + 2E [f (b ′ _n (k) b ′ _n (k)
] + E [f (b ′ _n (k) ² ] 1-b ′ ² _{n + 1} (k)... 1-b ′ ² ₁ (k)] (48)
For example, when f (b ′ _n (k)) is _expressed by equation (46), D (k) is as follows.
D (k) = diag [1-b ′ ² ₁ (k) 1-b ′ ² ₂ (k)...
1-b ′ ² _n−1 (k) 1+ (2α + α ² ) b ′ ² _n (K) 1-b ′ ² _{n + 1} (k)... 1-b ′ ² ₁ (k)] (49)
An example of a functional configuration for estimating the adaptive filter coefficient w _n (k) when the error correction decoding result is reflected in the signal to be detected in this way is a transmission signal b ₁ (k from the first transmitter as a signal to be detected. ) Is shown in FIG. 7A. The soft decision transmission symbol b ′ ₁ (k) is input to the function calculation unit 331-1 and the function calculation f (b ′ ₁ (k)) is calculated. The soft decision transmission symbols b ′ ₁ (k) to b ′ _N (k) and f (b ′ ₁ (k)) from the _N decoders are input to the error matrix generation unit 332-1 and the equation (35) ), Expression (36), and Expression (48), the error matrix G (k) is calculated and generated. The error matrix G (k), the estimated channel matrix H, and the noise power σ ² are input to the filter coefficient generation unit 333-1, where equation (34) is calculated, and the adaptive filter coefficient w _n (k) is calculated. Presumed. In this case, f (b ′ _n (k)) is also input to the interference replica vector generation unit 314-1, and the interference replica vector B ′ (k) of Equation (29) from Equation (30) and Equation (43) is obtained. Generated. The differential vector y ′ (k) is filtered by the adaptive filter unit 318-1 with the filter coefficient w _n (k) to obtain the log likelihood ratio Λ ₁ [b ₁ (k)]. In the case of the filter coefficient estimation unit 317-1 in FIG. 2, the function calculation unit 331-1 in FIG. 7A is omitted, and only the soft decision transmission symbols b ′ ₁ (k) to b ′ _N (k) are error matrices. The data is input to the generation unit 332-1 and the expression (34) is calculated.

In FIG. 3, an interference replica vector B ′ (k) is generated in step S4, and steps S5 to S7 are further processed to obtain a filter coefficient w _n (k) in step S8. When calculating (34), as shown in FIG. 7B, using the soft decision transmission symbols b ′ ₁ (k) to b ′ _N (k) in step S8-2, equations (35) to (37) are used. To generate an error matrix G (k), and in step S8-3, using the error matrix G (k), the estimated channel matrix H, and the noise power σ ² , the adaptive filter coefficient w _n is calculated by the calculation of Expression (34). Find (k).

When it is desired to reflect the error correction decoding result in the signal detected as described above, in FIG. 7B, the soft decision transmission symbol b ′ _n (k) of the signal desired to be detected in step S8-1 before step S4 is subjected to a function calculation. Using this, in step S4, the equation (43) is used instead of the equation (31), that is, the interference replica vector B ′ (k) is generated by the equations (29), (30), and (43). In step S8-2, equation (48) may be used instead of equation (37). As described above, when f (b ′ _n (k) is αb ′ _n (k) or αb ′ _n (k) ² and α is changed, the number of times of processing or decoding is performed in step S8-1-1. Α may be determined based on the reliability of the entire frame, and 1+ (2α + α ² ) b ′ _n (k) ² may be calculated and used as f (b ′ _n (k)) in step S8-1-2.

This technique of reflecting the error correction result in the detected signal can also be applied to the single user turbo receiver described in the section of the prior art. Further, in the method of reflecting the error correction result in the signal to be detected, the approximation shown in the equations (39) and (40) can be applied, and in this case, the filter coefficient is shown as shown by a broken line in FIG. 7A. It suffices to input only the matrix h shown in Expression (38) from the channel estimator 28 to the generation unit 333-1.
In the above description, the adaptive filter coefficient w _n (k) is obtained by the equation (34), that is, obtained using the channel matrix H, but the channel matrix H may not be used. That is, in the first decoding process (turbo reception process), the error vector G in Equation (34) is a unit matrix. Therefore, the filter coefficient generation unit 333 generates the difference vector y ′ (k) and the training signal or the hard decision transmission symbol b ^ _n (k), preferably b ^ _n (k) having high reliability as described above. The adaptive filter coefficient w _n (k) may be sequentially calculated by inputting −1 and applying RLS (recursive least squares) or the like. Since the error vector G depends on the discrete time k, it is necessary to update the adaptive filter coefficient w _n (k) for each symbol after the second iteration of the decoding process. As described above, the channel matrix H Is preferably used to determine the adaptive filter coefficient w _n (k).

Fourth invention (channel estimation )
As described above, it is possible to use not only known information such as a unique word but also a hard decision value of an information symbol, particularly a probable one as a reference signal for iterative channel estimation, which is used in the multi-input multi-output turbo reception method. In general, the channel (transmission path) of the received signal is estimated from the received signal and the known signal, and the received signal is processed and decoded using the estimated channel value. Can be applied to a turbo reception method in which the same received signal is repeated and processing based on the estimated channel value and decoding processing are performed.

  FIG. 8 shows an example in which the hard decision value of this information symbol is also applied to the channel estimation and turbo equalizer 41. The turbo equalizer 41 determines a linear equalization filter coefficient based on the estimated channel value, processes the received signal using the linear equalization filter, decodes the processed signal, and repeats the same received signal using the decoded signal. To process. The received signal r (k) is input to the turbo equalizer 41 and supplied to the channel estimator 42. The channel estimator 42 uses the received signal r (k) and the unique word from the storage unit 29 to determine the channel value (transmission path). Characteristic) is estimated, and the received signal r (k) is equalized in the turbo equalizer 41 based on the estimated channel value, then decoded, and decoded data c ′ (i) is output. Determination value b ′ (i) is output. If the soft decision value b '(i) is input to the symbol selector 43 and the absolute value of the soft decision value b' (i) is greater than or equal to the threshold Th, the hard decision value b ^ (i) In the channel estimation process in the channel estimation unit 42 when it is updated and stored in the previous symbol storage unit 32 as it seems (highly reliable) and the same received signal r (k) is repeatedly received (equalized) thereafter, Not only the unique word but also the hard decision value b ^ (i) of the information symbol stored in the previous symbol storage unit 32 is used.

The turbo equalizer 41 is, for example, a portion excluding the repetitive channel estimator 28, the unique word storage unit 29, and the previous symbol storage unit 32 in the receiver shown in FIG. The receiver in FIG. 31 may be used. That is, also in this case, the solution of the equation (19) becomes the following by the Wiener solution.
w (k) = E [y ′ (k) y ′ ^H (k)] · E [b (k) · y ′ (k)]
= [HΛ (k) H + σ ² I] · h (50)
Where H is defined by equation (8),
h≡ [H (Q−1),..., H (0)] ^T
H () is defined by equation (5), σ ² = E [‖v‖ ² ] (noise variance)
Λ (k) = diag [1−b ′ ² (k + Q−1),..., 1, 1−b ′ ² (k−Q + 1)]
As described above, the receiver in FIG. 31 also estimates the channel H (), obtains the equalization filter coefficient w (k) using this channel H (), and filters the received signal with the filter coefficient w (k). Then, the decoding processing is performed on the processed output. Therefore, in this iterative reception process, more accurate channel estimation can be obtained by using the reliable hard decision information symbol for channel estimation.

FIG. 9 shows an example of a turbo receiver in which the repetitive channel estimation method is applied to repetitive reception in which rake combining processing is performed. The received signal r (k) is supplied to the RAKE combining processor 45 and the channel estimator 42. In the first time, the channel estimator 42 estimates the channel value from the received signal r (k) and the unique word, and the RAKE combining processing unit 45 performs compensation for the phase rotation received by each symbol on the transmission path and RAKE combining processing. This is performed based on the estimated channel value, that is, time diversity processing is performed and output to the turbo decoder 46. The turbo decoder 46 outputs decoded data c ′ (i) and a soft decision value b ′ (i). The soft decision value b ′ (i) is input to the symbol selector 43, and the hard decision value b ^ (i) of the probable information symbol is updated and stored in the previous symbol storage unit 32 as in the above example.
In the second and subsequent RAKE reception-turbo decoding repetitive reception processing, the channel estimator 42 uses not only the unique word but also the hard decision value of the previous information symbol for channel estimation. Thereby, since channel estimation can be performed more accurately, quality can be improved.

FIG. 10 shows an example of a turbo receiver in which the repetitive channel estimation method is applied to repetitive reception using an adaptive (adaptive) array antenna. The received signal r (k) is received by the adaptive array antenna receiver 47, and the received signal is branched and input to the channel estimator 42, and channel estimation is performed using this and the unique word, and the estimated channel value is used. The array weight determining unit 48 sets each antenna element or corresponding reception path so that the main beam of the antenna directivity of the adaptive array antenna receiving unit 47 is directed in the arrival direction of the target wave and null is directed in the arrival direction of the interference wave. The weight for is determined, and the weight is set at the corresponding location. The reception output of the adaptive array antenna receiving unit 47 is supplied to the turbo decoder 46 and decoded, and the decoded data c ′ (i) and the soft decision value b ′ (i) are outputted. The soft decision value b
′ (I) is input to the symbol selector 43, and a probable hard decision value is updated and stored in the previous symbol storage unit 32. In the second and subsequent adaptive array antenna reception unit 47-repeated reception processing by the turbo decoder 46, the channel estimator 42 uses not only the unique word but also the hard decision value of the previous information symbol for channel estimation. As a result, channel estimation is performed more correctly, and as a result, the antenna directivity is controlled more accurately and quality can be improved.

  The turbo equalizer 41 in FIG. 8 is simply shown as a serial connection form of a soft input / soft output (SISO) equalizer (equalizer) 41a and a SISO decoder (decoder) 41b as shown in FIG. 11A. An iterative operation is performed between the equalizer 41a and the decoder 41b. The turbo decoder 46 in FIG. 9 and FIG. 10 is simply shown in FIG. 11B in the form of serial connection of the SISO decoder 46a and the SISO decoder 46b, and iterative decoding is performed between the decoders 46a and 46b. The turbo decoder 46 in FIGS. 9 and 10 may be a single SISO decoder.

The examples shown in FIGS. 8 to 10 are collectively shown in FIG. That is, the received signal is repeatedly processed by the receiver (turbo receiver) 49 using the channel value estimated by the channel estimator 42, the processed signal is decoded, and the decoded data (symbol) c ′ is obtained as a result of the decoding process. (I) and its soft decision value b '(i) are output, and the soft decision value b' (i) is compared with the threshold value in the symbol selector 43, and the corresponding decoded data c '(i) ( (Symbol hard decision value) is determined to be probable, and if it is determined to be probable, the hard decision value is updated and stored in the previous symbol storage unit 32, and the process using the estimated channel value after the second time is performed. The channel estimation in the channel estimator 42 in the repetition of the decoding process is performed more accurately by using the previous symbol hard decision value in addition to the known information such as a unique word. It is intended.

  FIG. 13 shows an example of the processing procedure of the iterative reception signal method that also uses this symbol hard decision value. In step S1, a channel value is estimated from the received signal and the known signal, and in step S2, it is checked whether or not it is the first iteration. In the first time, the received signal is processed in step S3 using the channel value estimated in step S1. Thereafter, a decoding process is performed to obtain a symbol hard decision value and a soft decision value. In step S4, a probable symbol hard decision value is extracted from the symbol soft decision value, and in step S5, the previous symbol hard decision stored in the storage unit 32 is updated to the extracted symbol hard decision value. In step S6, it is checked whether the decoding process has been performed a predetermined number of times. If the predetermined number of times has not been reached, the process returns to step S1. If it is not the first iterative process in step S2, the previous symbol hard decision value is read from the storage unit 32 in step S7, channel estimation is performed based on this and the information symbol of the received signal, and the process proceeds to step S3.

Also in this case, as described in steps S1 ′ to S4 ′ with reference to FIG. 6, the second and subsequent processing may not use a known signal.
In the example shown in FIG. 10, the RAKE combining processing unit 45 may be inserted between the adaptive array antenna receiving unit 47 and the turbo decoder 46 as indicated by a broken line. In this case, channel estimation for each symbol phase rotation correction and RAKE combining in the RAKE combining processing unit 45 may be shared by the channel estimator 42 or may be provided individually.

Noise other than white Gaussian noise Example of turbo reception method (first invention) described above, example of second invention considering error correction, and turbo reception method (fourth invention) characterized by channel estimation method In the embodiment, the processing is performed assuming that the noise is white Gaussian noise. That is, it is assumed that v _m (k) in the right side of Expression (20) indicating the received signal r _m (k) of each antenna is white Gaussian noise. Here, white Gaussian noise follows a Gaussian distribution,
E [v _m (k) · v _m (k−q)] = σ ² : When q = 0, 0: When q ≠ 0 E [] is an expected value, and σ ² is a variance value.
Is a signal having statistical properties. An example of white Gaussian noise is thermal noise generated in an antenna element. The assumption of the white Gaussian noise is reflected in the part of σ ² I in the equation (34) for obtaining the filter coefficient w _n (k) or the equation (50) for obtaining the filter coefficient w (k). For example, w _n (k) in equation (34) is
w _n (k) = (HG (k) H ^H + E [n (k) · n ^H (k)]) ⁻¹ h
= (HG (k) H ^H + σ ² I) ⁻¹ h
It is calculated through the process. Here, E [n (k) · n ^H (k)] = σ ² I is calculated on the assumption that v _m (k) is white Gaussian noise having variance σ ² . The channel matrix H estimated by the iterative channel estimator 28 (FIG. 1) or 42 (FIG. 12), σ ^2, and the error matrix G (k) calculated from the prior log likelihood values are expressed by the equation (34). ) To calculate the filter coefficient w _n (k).

Consider the case where the noise v _m (k) is not white Gaussian noise. In this case, since E [n (k) · n ^H (k)] = σ ² I cannot be obtained, the expected value (covariance) of the noise component is used to calculate the filter coefficient w _n (k). It is necessary to estimate the matrix E [n (k) · n ^H (k)] by another method. This method will be described below. Here, the covariance matrix of the noise component is abbreviated as U≡E [n (k) · n ^H (k)]. When y (k) = H · B (k) + n (k) in equation (22) is transformed to n (k) = y (k) −H · B (k) and substituted into the covariance matrix U, It becomes an expression.

U = E [n (k) · n ^H (k)]
= E [(y (k) −H · B (k)) · (y (k) −H · B (k)) ^H ]
If the vector y (k) by the received signal, the estimated value H ^ of the channel matrix H by the channel estimation value, and B (k) by the reference signal are available, the matrix U is obtained by the time averaging method.
U ^ = Σk _{= 0} ^Tr (y (k) -H ^ B (k)) * (y (k) -H ^ B (k)) ^H (51)
Can be estimated. Here, Tr is the number of reference signal symbols.

The covariance matrix U ^ is estimated using the equation (51) together with the channel matrix H during the iterative channel estimation in the iterative channel estimator 28 or 42. The procedure is shown in FIG. FIG. 14A shows unique words and information symbol sequences in one frame in the received signal, and FIG. 14B shows the first and subsequent processing. In the first process, only the unique word is used as a reference signal, and the channel matrix H is first estimated.
Next, U is estimated by the formula (51) using the unique word and its channel matrix estimation value H ^. Using these estimated values U and H ^, the filter coefficient w _n (k)
w _n (k) = (H ^ G (k) H ^ ^H + U ^) ^-1 h (52)
And the received signal is equalized for the first time using the filter coefficient w _n (k) to estimate the transmission information symbol.

In the second process, H is set in the same procedure as the first process using both the unique word and the information symbol estimated by the threshold among the information symbols estimated in the first equalization * as reference signals. After re-estimation, U is re-estimated. By repeating this operation, the channel matrix estimate H ^ becomes more accurate with each iteration, the U estimate becomes more accurate, the accuracy of the filter coefficient w _n (k) increases, and the equalizer Improved characteristics.
Through the above processing, turbo reception can be performed when noise that is not white Gaussian noise is included in the received signal.

The functional configuration in the case of performing the linear equalization process by estimating the noise covariance matrix U in the received signal described above is the same as that from the first transmitter of the multi-output equalizer 31 shown in FIG. FIG. 15 shows an example applied to obtaining the log likelihood ratio Λ ₁ [b ₁ (k)] as the equalized output of the received signal of the transmitted signal. In FIG. 15, the same reference numerals are assigned to the portions corresponding to FIG.
A unique word from the unique word storage unit 29 or a probable previous symbol hard decision from the previous symbol storage unit 32 is input to the reference vector generation unit 319, where the reference vector B (k ) Is generated. This reference vector B (k), the estimated channel matrix H ^ from the channel estimator 28, and the received vector y (k) from the received vector generating unit 311 are supplied to the covariance matrix estimating unit 321, where 51) is calculated to obtain an estimation matrix U ^ of the covariance matrix U.

Also, the soft decision transmission symbol soft decisions b ′ ₁ (k) to b ′ _n (k) from the soft decision symbol generation unit 313-1 are input to the error vector generation unit 322-1. Here, the equations (35) and (35) (36) and Equation (37) generate an error matrix G ₁ (k) corresponding to the channel estimation square error. The error matrix G ₁ (k), the estimated covariance matrix U ^, and the estimated channel matrix H ^ are supplied to the filter estimator 323-1, where equation (52) is calculated and the filter coefficient w ₁ ( k) is estimated. The filter coefficient w ₁ (k) and the difference vector y ′ (k) from the difference calculation unit 316-1 are supplied to the adaptive filter 318-1, and the filter processing w ₁ (k) ^H y for y ′ (k) is performed. ′ (K) is made and the result is output as the log likelihood ratio Λ ₁ [b ₁ (k)].

When the error correction decoding result is also reflected in the signal to be detected, as shown by a broken line in FIG. 15, the function calculation unit 331-1 shown in FIG. 7A is provided to calculate f (b ′ _n (k)). The interference replica vector generation unit 314-1 may use the equation (43) instead of the equation (31), and the error vector generation unit 322-1 may use the equation (48) instead of the equation (37).
The technique shown in FIG. 14B is shown as a flowchart in FIG. That is, in step S1, the channel matrix H is estimated using the received signal r (k) and a known signal (for example, a unique word). Next, in step S2, whether or not this process is the first in the iterative process is checked. If it is the first time, the estimated covariance matrix U ^ is obtained by calculating Equation (51) using the known signal, the estimated channel matrix H ^ and the received signal r (k) in step S3.

In step S4, the equation (52) is calculated using the estimated channel matrix H ^, the estimated covariance matrix U ^, and the error matrix G (k) composed of symbol soft decision values, and the filter coefficient w _n (k) is obtained. presume.
In step S5, the received signal is equalized using the estimated channel matrix H ^ and the filter coefficient w _n (k), that is, the equation (27) is calculated, and w _n ^H (k) · y ′ (k) is calculated. Then, a log likelihood ratio Λ ₁ [b _n (k)] is obtained, and decoding processing is performed on the log likelihood ratio Λ ₁ [b _n (k)] to estimate the hard decision value and the soft decision value of the transmission symbol.

In step S6, a probable (highly reliable) symbol hard decision value corresponding to the symbol soft decision value equal to or greater than the threshold is obtained. The symbol hard decision value stored in the previous symbol storage unit 32 is updated with the symbol hard decision value. Thereafter, in step S8, it is checked whether the number of decoding processes has reached a predetermined value. If not, the process returns to step S1, and if it has reached the predetermined value, the process for the received frame is terminated.
If the process in the iterative process is not the first time in step S2, that is, if it is after the second time, the symbol hard decision value is read from the previous symbol storage unit 32 in step S9, and the channel is determined by this and the information symbol in the received signal. The matrix H is estimated and the process proceeds to step S3.

Also in this case, the known signal can be prevented from being used after the second time by changing the steps S1 and S2 to the same processing as the steps S1 ′ to S4 ′ indicated by the broken lines in FIG. Further, when it is desired to reflect the error correction decoding result in the signal to be detected, the function calculation f (b ′ _n (k)) is performed in step S10 as shown by the broken line in FIG. k) may be obtained. Further, in any case, it is not necessary to use the hard decision transmission symbol for estimating the covariance matrix U ^.
The ability to estimate the noise covariance matrix U in a received signal including noise that is not white Gaussian noise can be applied to various useful applications as described below.

(1) A reception method for a multi-sequence transmission signal in which an interference signal whose receiver is unknown is included. As shown in FIG. 30A, in addition to the N series of transmission signals, such as signals from the transmitters of N users to be received by the turbo receiver, the turbo receiver is unknown at the turbo receiver as indicated by a broken line. Assume that an interference signal i (k) (for example, a signal from another cell or zone in mobile communication) is received. At this time, the equation (20) becomes
r _m (k) = Σ _{q = 0} ^Q−1 Σ _{n = 1} ^N h _mn (q) · b _n (k−q + 1) + i
(K) + v _m (k) (20) ′
It becomes. In this model, if i (k) + v _m (k) ≡v ′ _m (k),
r _m (k) = Σ _{q = 0} ^Q−1 Σ _{n = 1} ^N h _mn (q) · b _n (k−q + 1) + v
′ _M (k) (20) ″
It becomes. v ′ _m (k) is a noise signal that is not white Gaussian noise, H estimation and U estimation are performed as described above, w _n (k) is estimated, received signal equalization processing, and transmission Turbo reception can be performed by repeating symbol estimation.

(2) In a communication system using a transmission / reception separation filter, when oversampling a received signal at a speed higher than half the symbol period, a noise component included in the received signal sampled at each time There is a correlation between them, and the noise in the received signal cannot be regarded as white Gaussian noise. That is, in equation (20),
E [v _m (k) · v _m (k−q)] = σ ² : When q = 0, 0: q ≠ 0 is not the case. Therefore, E [n (k) · n ^H (k)] = σ ² I
It cannot be assumed. Therefore, by processing the received signal separated by the transmission / reception separation filter by obtaining the covariance matrix U using Equation (51), the received signal can be processed correctly.

  (3) In the turbo reception method described above, all Q path multipath components from each transmitter (user) are combined. However, when there is a long delay wave in the channel (eg, a path component of 1 symbol delay, 2 symbol delay, 3 symbol delay, 30 symbol delay when there is a 30 symbol delay), a long delay It is possible to adopt a policy that does not synthesize waves, treats them as unknown interference, and removes them with an adaptive filter. That is, by treating this long delayed wave component as the interference signal i (k) in the example (1), it is possible to remove the long delayed wave.

In the processing for the received signal including noise that is not white Gaussian noise as described above, the estimation of the covariance matrix U can be applied to the single user turbo reception method by estimating instead of σ ² I in Equation (50), Similarly, regardless of single-user or multi-user, the channel estimator 42 in the RAKE combining process reception shown in FIG. 9 or the turbo reception using the adaptive array antenna reception shown in FIG. 10, and more generally the iterative decoding shown in FIG. Can be applied to channel estimation and covariance matrix U estimation. In the case of RAKE reception, only channel estimation may be used.

Third invention (multi-stage equalization)
In the above description, the received signals r ₁ ,..., R _M are equalized by the multi-output equalizer 31 to obtain the log likelihood ratios Λ ₁ [b (k)], ..., Λ _N [b (k)]. In the modification (2) of the first invention, a plurality of equalization stages may be provided in cascade, and the number of outputs may be reduced as the subsequent equalizer. For example, as shown in FIG. 17, the former stage equalizer (multi-user equalizer) 71 cancels the interference component outside the equalization range of the latter stage single user equalizer 21 ′. Pre-processing of soft interference cancellation and MMSE (minimum mean square error) normative linear filtering is performed, and then equalization processing of a single user whose number of paths is Q is performed by the post-stage equalizer 21 ′.

In this way, it is possible to prevent the amount of calculation from becoming enormous by performing cascade equalization processing and using a linear filter in the previous processing.
FIG. 18 shows a configuration example of a multi-output turbo receiver of an embodiment based on the basic concept of the first invention (2) of the turbo reception method and a configuration example of a MIMO system to which the present invention is applied. The same reference numerals are assigned to the corresponding parts, and redundant description is omitted (the same applies to the following description).
A transmission signal from each transmitter is received by the turbo receiver 30 through the transmission path (channel). The received signal r (k) is input to the multi-user equalizer 71, from which the signals from each of the N transmitters are eliminated by interference from signals from other transmitters. The signals u ₁ (k),..., U _N (k) and the channel values α ₁ (k),..., Α _N (k) are output and single-user equalizers 21-1,. ,..., 21-N output log likelihood ratios Λ ₁ [b ₁ (k)],..., Λ ₁ [b _N (k)], respectively. The subsequent processing is the same as in FIG. 1, but the channel values α ₁ (k),..., Α _N (k) used in the single user equalizers 21-1,. This is a channel value after user equalization and is different from the channel matrix H. Therefore, α ₁ (k),..., Α _N (k) are referred to as channel information after equalization.

Hereinafter, the operation of each unit will be described.
In consideration of the number Q of multipaths (channels), equations (23) to (26) are defined in the same manner as in the description of FIG.
The downstream equalizers 21-1,..., 21-N in FIG. 18 have their own signal symbols [b _n (k), b _n (k−1),..., B _n (K−Q + 1). ] (N = 1,..., N) equalize the intersymbol interference channel. Therefore, the equalizer 71 in the previous stage is other than [b _n (k), b _n (k−1),..., B _n (K−Q + 1)] (n = 1,..., N) in y (k). To remove the interference. The quantitative explanation is given below.

First, soft decision transmission symbols using the prior information λ ₂ ^p [b _n (k)] (n = 1,..., N) of the equalizer 71 fed back from the decoders 24-1,. Estimate b '(k) is obtained by equation (15).
Next, a replica H · B ′ (k) of the interference signal is created using these soft decision transmission symbols b ′ _n (k) and the channel matrix H, and is subtracted from the received vector y (k).
y ′ _n (k) ≡y (k) −H · B ′ (k) (27) ′
= H. (B (k) -B '(k)) + n (k) (28)'
here,
B ′ (k) = [b ′ ^T (k + Q−1)... B ′ ^T (k)... B ′ ^T (k−Q + 1)] ^T (29) ′
And
b ′ (k + q) = [b ′ ₁ (k + q) b ′ ₂ (k + q)... b ′ _n (k + q)... b ′ _N (k + q)] ^T : q = Q−1,.
b '(k + q) = [b' 1 (k + q) b '2 (k + q) ... 0 ... b' N (k + q)] T: q = 0, ..., -Q + 1 (54)
(The zero in the element of b '(k + q) is nth)
Hereinafter, the operation of subtracting this interference will be referred to as soft interference cancellation. Assuming that a replica of the interference signal is ideally created, y ′ _n (k) obtained after subtraction is the symbol b _n (k) of the n-th user and q = 1, ..., -Q + 1, based on the fact that the nth element of b '(k + q) is set to 0, the nth user's own symbol [b _n (k-1), ..., b _n (k-Q + 1) It can be seen that it can only have an intersymbol interference component.

The contribution component from the signal of the nth user (transmitter) in the actual reception vector r (k) is the symbol [b _n (k), b _n (k−1),..., B _n (k−Q + 1)]. However, as can be understood from the definition of the reception vector y (k) in the equation (21), the nth user (transmitter) in the reception vector y (k) generated by combining the multipaths. If the k-th symbol b _n (k) is used as a contribution component from the signal, future symbols [b _n (k + Q−1), b _n (k + Q−2),..., B _n ( k + 1)] is also included. That is, the interference replica includes interference components from the future. As described above, the difference vector y ′ (k) in Expression (27) ′ is different from the difference vector y ′ (k) in Expression (27).

Therefore, the next step of the pre-processing in the equalizer 71 is an interference surplus component after soft interference cancellation, that is, a residual interference component based on incomplete synthesis of the interference replica H · B ′ (k), and the future intersymbol interference component. Is removed from y ′ _n (k) by a linear filter of the MMSE (Minimum Mean Square Error) criterion. That is, the y _'n (k) by the filter characteristic w _n, a result of the filter processing as shown in equation (55) is a symbol in the signal of the n-th user in the received signal [b _n (k), b _n (k−1),..., b _n (K−Q + 1)] are multiplied by channel values α _1n , α _{2n ,.}
w _n ^H (k) · y ′ _n (k) ≈Σq _{= 0} ^Q−1 α _q (k) · b _n (k−q) = α _n ^H (k) · b _n (k) (55)
Therefore, the filter characteristic w _n (k) and the equalized channel value (channel information) α _n (k) may be obtained to calculate the equation (55). The calculation method of w _n (k) and α _n (k) is shown below. Although the filter characteristic w _n (k) is different from the filter coefficient w _n (k) given by the equations (32) and (34), the same symbol is used for convenience.

The above solution is defined as the solution of the following optimal problem.
_{(W n (k), α} n (k)) = arg min ‖w n H (k) · y 'n (k) -α n H (k) · b n (k) || ² (56)
_{Assuming that} α _1n (k) = 1.
That is, w _n (k) and α _n (k) that minimize the right side of Expression (56) are obtained.
The added constraint condition α _1n (k) = 1 is to avoid a solution of α _n (k) = 0, w _n (k) = 0. this is,
‖Α _n (k) ‖ ² = 1
Although it is possible to solve with the following constraint condition, a solution in the case of α _1n (k) = 1 is shown below. For simplicity, replace the problem as follows: That is, the right side of the equation (56) is defined as m _n (k) that minimizes w and α.

m _n (k) = arg min ‖m _n ^H (k) · z _n (k) ‖ ² (57)
The condition is _mn ^H (k) · e _{MQ + 1} = −1. (Equivalent to α _1n (k) = 1)
here,
m _n (k) ≡ [w _n ^T (k), −α _n (k) ^T ] ^T (58)
z _n (k) ≡ [y _n ^T (k), b (k) _n ^T ] ^T (59)
e _{MQ + 1} = [0 ... 1 ... 0] ^T (60)
(1 element in e _{MQ + 1} is MQ + 1)
It is. Reference [2] S. Haykin, Adaptive Filter Theory, Prentice Hall P.220〜P
From the Lagrange undetermined coefficient method shown in 227, the solution to this optimization problem is given by

m _n (k) = − R _ZZ ⁻¹ · e _{MQ + 1} / (e _{MQ + 1} ^H · R _ZZ ⁻¹ · e _{MQ + 1} ) (61)
here,
R _ZZ = Ε [z _n (k) · z _n ^H (k)] (62)
Ε [A] represents the expected value (average value) of A.

Λ _n (k) = diag [D _n (k + Q−1),..., D _n (k),..., D _n (k−Q + 1)] (64)
I is unit matrix σ ² is noise power (dispersion value of white Gaussian noise)

D _n (k + q) = diag [1-b ′ ₁ ² (k + q),..., 1-b ′ _n ² (k + q),..., 1-b ′ _N ² (k + q)] : Q = Q + 1, ..., 1 (66)
D _n (k + q) = diag [1-b ′ ₁ ² (k + q),..., 1,..., 1-b ′ _N ² (k + q)] : Q = 0, ..., -Q + 1 (67)
diag represents a diagonal matrix (elements other than the diagonal elements of the matrix are zero).
That is, if the channel matrix H, σ ² is known, m _n (k) can be obtained by the equation (61). Therefore, w _n (k) and α _n (k) are also obtained according to the equation (58).

Based on the calculated filter characteristic w _n (k), y ′ _n (k) is filtered by the following equation.
u _n (k) = w _n ^H (k) · y ′ _n (k) (68)
^H represents a conjugate transpose matrix.
The filtered n processing results are sent to the subsequent corresponding equalizer 21-n. In this way, the received signal u _n (k) corresponding to the left side of Equation (1) from the nth user is obtained, and α corresponding to the channel value h _mn (q) on the right side of Equation (1) is obtained. _mn (k) is obtained, that is, the equation (55) corresponding to the equation (1) is obtained. Therefore, α _n (k) is also given to the subsequent equalizer 21-n as an equalizer parameter (channel value). The preceding stage processing by the equalizer 71 has been described above.

Next, processing after the subsequent equalizer 21-n will be described. Since the expression (55) corresponds to the expression (1) as described above, the operation in the equalizer 21-n for each user may be performed in the same manner as the operation of the equalizer 21 in FIG. As described above, since it is shown in the document [1], details are omitted. Each equalizer 21-n receives u _n (k), α _n (k) defined above and a priori information λ ₂ [b _n (k)] from the decoder 24-n, and outputs each code as an output. The log likelihood ratio Λ ₁ (LLR: Log-Likelihood Ratio) of the probability that the quantization bit is +1 and the probability of −1 is calculated by the following equation.

Here, λ ₁ [b _n (k)] is external information sent to the subsequent decoder 24-n, and λ ₂ ^p [b _n (k)] is a priori information given to the equalizer 31. The decoder 24-n calculates the log likelihood ratio Λ ₂ by the following equation.

Here, λ ₂ [b _n (i)] is external information given to the equalizer 71 and the equalizer 21 in the repetition, and λ ₁ ^p [b _n (k)] is given to the decoder 24-n. It is prior information. With the configuration shown in FIG. 18, repeated equalization and decoding are performed to improve the error rate.
The functional configuration of the above-described multiuser equalizer 71 will be briefly described with reference to FIG. The received signal from each antenna is processed as a vector r (k) = [r ₁ (k)... R _M (k)] by the receiving unit 70, and each received vector generating unit 311 uses the vector r (k). A reception vector y (k) of Expression (21) considering multipath (channel) is generated.

On the other hand, the received signal r (k) from the receiving unit 70 and a known sequence signal such as a unique word sequence for channel estimation corresponding to each transmitter from the unique word storage unit 29 are input to the channel estimator 28. Thus, the channel matrix H is estimated.
Also, prior information λ ₁ ^p [b _{1 is obtained} from the output log likelihood ratios Λ ₂ [b ₁ (i)],..., Λ ₂ [b _N (i)] of the respective decoders 24-1,. , Λ ₁ ^p [b _N (i)] subtracted from the external information λ ₂ [b ₁ (k)], λ ₂ [b _N (k)] is a soft decision symbol estimator. , 313 -N, and soft decision transmission symbols b ′ ₁ (k),..., B ′ _N (k) are calculated by Expression (15), respectively, and these are input to the interference vector generation unit 72. The interference vector generation unit 72 generates a vector B ′ (k) of symbol estimation values that can be interference signals from other transmitters for each n by equations (29) ′, (53), and (54). The The products of these N vectors B ′ (k) and the channel matrix H are respectively calculated by other interference signal estimation units 73-1,..., 73-N to obtain interference component replicas H · B (k). .

These N interference component replicas H · B (k) are subtracted from the received vector y (k) by the subtracting units 74-1,..., 74-N, respectively, and the difference vector y ′ ₁ (k),. , Y ′ _N (k).
Soft-decision transmission symbols b ′ ₁ (k),..., B ′ _N (k) are input to the error matrix generation unit 75, and the error matrix Λ ₁ (k) is obtained by equations (64), (66), and (67). ,..., Λ _N (k) are generated, and the channel matrix H and noise power σ ² are input to the filter characteristic estimation unit 76. The filter characteristic estimation unit 76 uses equations (58), (60), (61). and (63) and (65), the filter characteristic w _n and channel information alpha _n after equalization is estimated. These filter characteristics w ₁ ,..., W _N and the difference vectors y ′ ₁ (k),..., Y ′ _N (k) are respectively multiplied by the filter processing units 77-1,. Then, interference from other user signals is removed from the received signals of symbols [b _n (k), b _n (k−1),..., B _n (K−Q + 1)] from each path for each user. u ₁ (k) is a component, ..., u _N (k) is obtained, respectively, the channel information after equalization obtained in these and the filter characteristic estimating unit _{76 α 1 (k), ...} , α N ( k) are supplied to single user equalizers 21-1,..., 21-N in FIG.

The processing procedure of the first invention (2) of this turbo reception method is shown in FIG. In FIG. 20, the same step symbols are assigned to steps corresponding to the processing procedure shown in FIG.
However, the calculation of the interference replica vector B ′ _n (k) in step S4 is performed by equations (29) ′, (53), and (54). In step S13, the soft decision transmission symbol b ′ _n (k) is used, and an error matrix Λ _n (k) is generated by equations (64), (66), and (67). The step S14 uses the channel, the matrix H, the noise power σ ² and the error matrix Λ _n (k), and the residual interference removal filter w _{n according} to the equations (58), (60), (61), (63), (65). (K) and channel information α _n are obtained.
Step S15 obtains the u _n (k) and the difference vector y _'n (k) is filtered by the residual interference elimination filter characteristic w _n (k) in. In step S16, each filter processing result u _n (k) is subjected to single user equalization processing to obtain log likelihood ratios Λ _n [b _n (k)], and these are decoded in step S10. The other processes are the same as those shown in FIG.

In the above description, the equalization range in the post-stage equalizer 21-n is the symbol [b _N (k), b _n (k−1),..., B _n (K−Q + 1)] (n = 1,..., N). However, this equalization range can be adjusted. For example, when Q is a very large value, the calculation load of the subsequent equalizer 21-n increases. In such a case, the equalization range in the post-stage equalizer 21-n is Q ′ <Q, and the pre-stage equalizer 71 uses [b _n (k), b _n (k−1),. _n (K−Q ′ + 1)] (Q ′ <Q, n = 1,..., N) may be changed so as to eliminate intersymbol interference of signals of the same user other than the section. This change will be described later. Sometimes used separately in the front equalization and subsequent equalization, also used in the channel estimator 28 is provided previous symbol memory 32 hard decision transmitted symbol b ^ _n (k) as shown by a broken line in FIG. 19 Thus, the estimation accuracy can be improved by estimating the channel value.

To the transmission signal of the preceding stage in the multi-output equalizer 71 N series in the example shown in FIG. 17, they were equalized separated interference from other sequences, a signal u _n of N series channel after equalization The information α _n is output, and then the intersymbol interference of the same transmission signal is removed from each N-sequence signal u _n by the single-user equalizer 22-n at the subsequent stage. That is, a two-stage cascade equalization configuration is adopted. It is good also as a cascaded multistage structure of three or more stages.
For example, as shown in FIG. 21, the equalizer 81 of the first stage, the U + 1 th transmission sequence of the first to U-th transmission sequence to input the received signal r _m of the M-sequence to the transmission signal of N series The interference by the equalized signal sequence er ₁ (k) from which the interference due to Eq. And the equalized channel information eα (k) and the 1st to Uth transmission sequences of the (U + 1) th to (N) th transmission sequences are eliminated The equalized signal sequence er ₂ (k) and the channel information eα ₂ (k) after the equalization are obtained and input to the second stage equalizers 82-1 and 82-2 at 82-1. was er ₁ (k) and eα ₁ (k) and the equalization processing, according to U ₁ + 1 the U th transmission sequence of the first to U-th first through U ₁ th transmission sequences in transmission sequences interfere with the equalized signal to remove sequences er ₃ (k) and the channel information Iarufa ₃ after the equalization (k), the first of the first to during the U-th transmission sequence ₁ + 1 first to U ₁ No. transmission sequences and the U ₂ ~ equalized signal sequence er ₄ removing the interference due to the U-th transmission sequence (k) and the channel after equalization that of the U ₂ No. transmission sequences Information eα ₄ (k), an equalized signal sequence er ₅ (k) from which interference due to the first to U ₂ transmission sequences of the U ₂ +1 to U transmission sequences in the first to U transmission sequences is removed, and The equalized channel information eα ₅ (k) is output.

Similarly, the equalization signal sequence er ₂ (k) and the channel information eα ₂ (k) are input to the second-stage equalizer 82-2, and the equalization signal sequence er ₆ (k) and the equalized signal sequence are obtained. The channel information eα ₆ (k), the equalized signal sequence er ₇ (k) and the equalized channel information eα ₇ (k) are output. In the case of N = 5, the third-stage equalizers 83-1 to 83-5 are single-user equalizers in FIG. Alternatively, the input equalization signal of the equalizer 83-3 is composed of two transmission signals, and the equalizer 83-3 eliminates the mutual interference between the two transmission signals, and two sets of equalization signals and The channel information after equalization may be equalized by the following single user equalizers 84-1 and 84-2. Further, for example, the equalizer 83-4 inputs the equalized signal er ₆ (k) and the channel information eα ₆ (k), and the other two of the constituent transmission signals, for example, each of the three transmission signals, are input. Mutual interference with the transmission signal and intersymbol interference due to its own multipath may be removed. One or more of the second-stage equalizers 82-1 and 82-2 may be configured to obtain the equalized signals for a plurality of transmission signals all at once.

As described above, generally, a plurality of equalized signal sequences and a set of equalized channel information are output from the first-stage equalizer, and each equalized signal sequence and a set of channel information after the equalization, One to a plurality of equalizers are cascaded in one to a plurality of stages, and finally the respective equalized outputs of the first to Nth transmission sequences, that is, the log likelihood ratio Λ ₁ [b _n ( k)] can also be output.
When multistage cascade equalization processing is performed in this way, it is preferable to reduce the amount of calculation processing by reducing the value of the number Q of paths from which interference is removed, as described above. In this case, as described above, the interference component due to the path reduced in the subsequent stage is removed in the equalization stage immediately before.

In the following, the equalizer 21 at the first stage in FIG. 21 performs equalization signal sequences of N transmission sequences, a group of U transmission sequences from reception signals with Q multipath numbers for each transmission sequence. er ₁ (k) and channel information eα ₁ (k) after equalization are obtained, and the number of multipaths in each transmission string system is Q ′ in the equalization processing in the equalizer 82-1 at the subsequent stage The equalization process will be described.
The interference vector generator 72 generates the interference vector B ′ (k) in substantially the same manner as in the embodiment shown in FIGS. 18 and 19. The constitutive equations (53) and (54) are expressed by the equations (53), (54) ′ and formula (73).

b ′ (k + q) = [b ′ ₁ (k + q) b ′ ₂ (k + q)... b ′ _n (k + q)... b ′ _N (k + q)] ^T : q = Q−1,.
b ′ (k + q) = [0... 0 b ′ _{U + 1} (k + q)... b ′ _N (k + q)] ^T : q = 0,..., −Q ′ + 1 (54) ′
b ′ (k + q) = [b ′ ₁ (k + q) b ′ ₂ (k + q)... b ′ _n (k + q)... b ′ _N (k + q)] ^T : q = Q ′,.
Equation (54) ′ is for equalization except for the symbols of the first to Uth transmission sequences themselves and the self and mutual intersymbol interference components of these sequences based on the Q ′ multipath. Since (73) reduces the number of multipaths to Q ′ by the subsequent equalization, it eliminates the self and mutual intersymbol interference of the first to U-th transmission sequences based on the Q ′ + 1 th to Q th paths. Is to do.

An interference signal replica H · B ′ (k) is created using the interference vector B ′ (k) obtained in this way, and is subtracted from the received vector y (k), that is, the following equation is calculated.
y ′ _g (k) ≡y (k) −H · B ′ (k) (27) ″
= H · (B (k) −B ′ (k)) + n (k) (28) ″
Hereinafter, the operation of subtracting this interference will be referred to as soft interference cancellation. Assuming that the replica H · B ′ (k) of the interference signal is ideally created, y ′ _g (k) obtained after subtraction is the symbol of the first to U-th transmission sequences, [b _n (k) , B _n (k−1),..., B _n (k−Q ′ + 1)], (n = 1 to U).

Next, the excessive interference component after the soft interference cancellation is removed by the MMSE standard linear filter in the same manner as described above. An expression corresponding to the expression (55) in this case is the following expression (55) ′.
w _g ^H (k) · y ′ _g (k) ≈Σ _{n = 1} ^U Σ _{q = 0} ^Q′-1 α _nq (k) · b _n (k−q) = α _g ^H (k) · b _g (K) (55) '
here,
α _g (k) = [α _1,0 (k), ..., α _{1, Q'-1} (k), ..., α _{U, 0} (k), ..., α _{U, Q'-1} (k) ] ^T (55-1)
b _g (k) = [b ₁ (k), ..., b ₁ (k-Q '+ 1), ..., b _U (k), ..., b _U (k-Q' + 1)] ^T (55-2 )
In order to obtain these w _g (k) and α _g (k), similarly to the above, w _g (k) and α _g (k) that minimize the right side are obtained using equation (56) as the following equation.

(W _g (k), α _g (k)) = arg min ‖w _g ^H (k) · y ′ _g (k) −α _g ^H (k) · b _g (k) ‖ ² (56) ′
The condition is α _1,0 (k) = 1.
The added constraint is to avoid a solution of α _g (k) = 0, w _g (k) = 0, and can be solved with a constraint of ‖α _g (k) ‖ ² = 1. However, when α _1,0 (k) = 1, the problem is replaced as follows.
_{m g (k) = arg min} ‖m g H (k) · z g (k) ‖ ² (57) '
The condition is _mg ^H (k) · e _{MQ ′ + 1} = −1.
here,
m _g (k) ≡ [w _g ^T (k), −α _g ^T (k)] ^T (58) ′
z _g (k) ≡ [y _g ^T (k), b (k) _g ^T ] ^T (59) ′
e _{MQ '+ 1} = [0 ... 1 ... 0] ^T (60)'
(1 element in e _{MQ '+ 1} is MQ' + 1)
From the Lagrange undetermined coefficient method shown in the literature [2], the solution of this optimization problem is given below.

m _g (k) = − R _zz ⁻¹ · e _{MQ ′ + 1} / (e _{MQ ′ + 1} ^H · R _zz ⁻¹ · e _{MQ ′ + 1} ) (61) ′
here,

Λ _n (k) = diag [D _n (k + Q−1),..., D _n (k),..., D _n (k−Q + 1)] (64) ′

D _n (k + q) = diag [1-b ′ ₁ ² (k + q),..., 1-b ′ _n ² (k + q),..., 1-b ′ _N ² (k + q)] : Q = Q + 1, ..., 1 (66)
D _n (k + q) = diag [1, ..., 1,1-b ′ _{U + 1} ² (k + q),..., 1-b ′ _N ² (k + q)] : Q = 0,..., −Q ′ + 1 (67) ′
D _n (k + q) = diag [1-b ′ ₁ ² (k + q),..., 1-b ′ _n ² (k + q),..., 1-b ′ _N ² (k + q)] : Q = Q ′,... −Q + 1 (74)
That is, if the channel parameter is known, _mg (k) can be obtained by Expression (61) ′. Further, w _g (k), α _g (k) (= eα ₁ (k)) is also obtained according to the equation (58) ′. Such calculation is performed by, for example, the filter characteristic estimation unit 76 in FIG. 19, and the filter processing unit 77-1 calculates the following expression and performs filter processing.

er ₁ (k) = w _g ^H (k) · y ′ _g (k)
The equalization output er ₁ (k) and post-equalization channel information eα ₁ (k) = α _g (k) are sent to the equalizer 82-1.
As described above, for example, when dividing 5 transmission sequences (users) into 3 transmission sequence (user) groups and 2 transmission sequence (user) groups, the above algorithm is executed with U = 3 and 2, Equalization outputs er ₁ (k), eα ₁ (k) and er ₂ (k), eα ₂ (k) are input to equalizers for the subsequent three transmission sequences (user) and two transmission sequences (user). Thus, an equalized output of each transmission sequence (user) is obtained.

Further, the error correction decoding result of the signal to be detected described above is reflected in the soft decision transmission symbol. The single user turbo equalizer receiver shown in FIG. 8, the RAKE combining process turbo receiver shown in FIG. The present invention can also be applied to a turbo receiver including an adaptive array antenna receiver, and more generally to a turbo receiver including the channel estimator 42 shown in FIG.
Further, in FIG. 13, FIG. 14 and FIG. 15, the symbol hard decision value determined to be probable for the second and subsequent estimations of the channel matrix H and the covariance matrix U ^ is also used as the reference signal. The covariance matrix U ^ may be estimated using only the word as a reference signal using the equation (51), and the channel estimation using the symbol hard decision value and the estimation of the covariance matrix U ^ may be omitted.
1st invention (2) (parallel transmission)
Next, it has been proposed to perform high-speed transmission with high frequency utilization efficiency by transmitting an information sequence c (i) by one user as a plurality of parallel sequences. An embodiment of a turbo receiver to which the present invention is applied to such a transmission signal will be described.

22, the same reference numerals are assigned to the portions corresponding to those in FIG. 1. On the transmission side, the modulation output signal b (j) from the modulator 13 is converted into N sequences by the serial-to-parallel converter 14. (J) is sequentially distributed to be an integer N sequence signals b ₁ (k),..., B _N (k) of 2 or more, which are not shown in the figure, but are converted into radio frequency signals. Later, it is transmitted from N antennas.
These N series signals are received by the turbo receiver of the present invention through a channel (transmission path). The receiver has one or more receiving antennas, and the received signal is an integer M or more baseband digital received signals r _m (k) (m = 1, 2,..., M) and is multi-output equalized. Is input to the device 31. The reception signal r _m (k) is generated, for example, as shown in FIG.

The multi-output equalizer 31 has the same configuration as that shown in FIG. 2, and performs processing similar to the processing procedure shown in FIG. At that time, the log-likelihood ratio lambda ₂ of from decoder 24 shown in FIG. 22 [b (i)] External information from lambda ₁ [bi ] Is subtracted by the subtracter 25, and the subtracted output is interleaved by the interleaver 26 to obtain the prior information λ ₂ [b (j)], and the prior information λ ₂ [b (j)] is converted into the serial-parallel converter 15 Are converted into N-sequence prior information λ ₂ [b ₁ (k)],..., Λ ₂ [b _N (k)] and input to the multi-output equalizer 31.

Therefore, in the multi-output equalizer 31, the received M-sequence signal is linearly equalized as described above, and N log likelihood ratio sequences Λ ₁ [b ₁ (k)],. ₁ [b _N (k)] is output. The N logarithmic likelihood ratio sequences are converted into one logarithmic likelihood ratio sequence Λ ₁ [b (j)] by the parallel-serial converter 16 and supplied to the subtractor 22. According to this configuration, the input signal format of the multi-output equalizer 31 is the same as that described with reference to FIGS. 1 to 3, and therefore, N series can be obtained by the equalization processing performed with reference to FIGS. 1 to 3. Log likelihood ratios Λ ₁ [b ₁ (k)],..., Λ ₁ [b _N (k)] can be obtained and repeated by using the serial-parallel converter 15 and the parallel-serial converter 16. It will be readily understood that the decryption process can be performed. Corresponding to the transmission signal of the nth transmitter in FIGS. 1 to 3, in this case, the nth (nth column) transmission signal in the N parallel transmission signals is equalized. In addition, it can be easily understood that the embodiment with reference to FIGS. 4 to 7 can be applied to reception for parallel transmission of this N-sequence signal. Also, the reception characteristics are improved by the cascade processing by the plurality of equalization stages shown in FIGS. 18 to 21 as compared with the process by the single equalization stage shown in FIGS.

The turbo reception method and receiver of the present invention can be applied to reception of convolutional code / turbo code + interleaver + multilevel modulation (QPSK, 8PSK, 16QAM, 64QAM, etc.), TCM (Trellis Coded Modulation) / turbo TCM, and the like.
Generation of M received signals In the above description, M received signals r ₁ (k),..., R _M (k) were obtained from M antennas # 1,. You may obtain | require, or you may obtain | require M received signals more than L from the received signal of the integer L antenna more than 2. Although the antennas # 1 did not specifically shown in FIG. 1, ..., received signal r ₁ of the baseband by the receiving signal baseband conversion unit from # M, ..., it is a r _m, digital discrete time k is sampled Signals r ₁ (k),..., R _M (k).

For example, as shown in FIG. 30B, the received signals received by L = 2 antennas # 1 and # 2 are converted into baseband signals by baseband converters 61-1 and 61-2, respectively. Each output of 61-1 and 61-2 is obtained by sampling signal from the sampling signal generator 62 and sampling signal obtained by shifting the phase of the sampling signal by T / 2 of the period T by the phase shifter 63, respectively. / D converters 64-1, 64-2 and 64-3, 64-4 sampled and converted to digital signals r ₁ (k), r ₂ (k), r ₃ (k), r ₄ (k) Then, it may be inputted to the turbo receiver 30 shown in FIG. 1, FIG. 18 or FIG. 22 to obtain N decoded outputs. Each sampling period of the received signals r ₁ (k),..., R ₄ (k) input to the turbo receiver 30 is a case where one received signal r _m (k) is received for each antenna. The frequency of the sampling signal from the sampling signal generator 62 is selected so as to coincide with the sampling period.

Experimental Example As described above, according to the first invention (1), a multiple output (MIMO) receiving method can be realized. FIG. 23 and FIG. 24 show error rate characteristics as quantitative effects. E _b / N _o of the horizontal axis in each figure is a bit energy to noise ratio. The following simulation conditions were assumed.
Number of users (transmitters) N 2
Number of multipaths Q 5 for each user
Number of receiving antennas 2 Number of information symbols in one frame 450 bits
Number of unique words in one frame 25 bits
Channel estimation method RLS (forgetting factor 0.99)
Error correction code rate 1/2, constraint length 3 convolutional code Doppler frequency 1000 Hz (Rayleigh fading)
Modulation method BPSK
Transmission speed 20Mbps
Decoder 24 Max-Log-Map decoder Number of repetitions: 4 No fading in frame Note that the approximation by the inverse theorem of the inverse matrix was not used for calculating the filter coefficient w.

FIG. 23 shows the error rate characteristics when channel estimation is completely performed (no estimation error), that is, the channel is known. The number of users (transmitters) N = 2, the number of receiving antennas M = 2, Rayleigh path number Q = 5. The first repetition is a state where the repetition is not performed, and is a result of performing the repetition once in the second repetition. It can be seen that the error rate characteristics are greatly improved by repetition.
This shows that the MIMO turbo reception method of the present invention operates properly.

FIG. 24 shows the effect of iterative channel estimation (fourth invention). The horizontal axis is the threshold value Th. E _b / N _o = 4 dB (E _b is for one user), and Th = 1.0, no symbol hard decision value is selected, that is, channel estimation using the symbol hard decision value is performed. Not considered a conventional method. In this case, as is apparent from the figure, the channel estimation is inaccurate, so that the repetitive effect of the BER characteristic is small. The threshold value Th = 0 is the case where all the hard decision values are used as they are, and if the hard decision values of the information symbols are also used in this way, the average bit error rate is improved as apparent from the figure, and the channel estimation is more accurate. It is understood that can be done. Furthermore, it can be seen that the average bit error rate is smaller at the threshold Th = 0.2 to 0.6 than when Th = 0, that is, it is better to use only the probable hard decision value. It is also understood that especially around Th = 0.25 is most preferable.

FIG. 25 shows the error rate characteristic of a MIMO reception method using a transmission symbol hard decision value that is more likely to be a threshold value for channel estimation, that is, using iterative channel estimation, as a curve 66. The result is a characteristic after 4 repetitions, N = 2, M = 2, Q = 5 Rayleigh, f _d T _s = 1 / 20000,900 symbols / frame. For comparison, the error rate characteristic when the channel estimation is perfect is shown in curve 67, and the hard decision value of the conventional information symbol is not used for channel estimation, that is, channel estimation without repetition (channel estimation is performed only once) is used. The error rate characteristic is shown in curve 68. From this graph, it can be seen that the error rate characteristic approaches that in the case of perfect channel estimation when channel repetition estimation is used.

Further, according to the channel estimation method described above, it is determined from the decoded soft decision value whether or not the hard decision value is probable, and the symbol information of the probable hard decision value is also obtained during the next iterative reception process. By using this for channel estimation, channel estimation can be performed more correctly and decoding quality can be improved.
Next, in order to confirm the effect of the embodiment in which the covariance matrix U ^ (noise other than Gaussian noise) is estimated, a simulation was performed under the following conditions.
Number of all users (transmitters) N 3 (including 1 user as unknown interference: i (k))
Number of multipaths Q 5 for each user
Number of receiving antennas 3 Number of information symbols in one frame 450 bits Error correction code rate 1/2, constraint length 3 convolutional code Doppler frequency 1000 Hz
Modulation method BPSK
Transmission speed 20Mbps
Decoder 24 Log-MAP is decoder repetition number 4 times 3 users (transmitters) have equal power. 26 shows simulation results of the BER (bit error rate) characteristics of the turbo receiver for estimating H and U ^ shown in FIGS. 14, 15, and 16. FIG. 27 shows the turbo receiver shown in FIG. The BER characteristics using the receiver using the above method as it is. In FIG. 26, only white Gaussian noise is used as the noise, and the effect is hardly obtained even if the channel estimation and decoding processes are repeated twice or more. In FIG. 27, the BER characteristic is improved by increasing the number of repetitions. It is understood that the BER is considerably smaller than that shown in FIG. 26 for the same Eb / No.

Next, in order to confirm the effect of the embodiment (second invention) in which the error correction decoding result is reflected on the symbol soft decision value b ′ _n (k) of the received signal from the target user (transmitter), The simulation was performed under the following conditions.
Number of all users (transmitters) N 4
Number of multipaths Q 5 for each user
Number of receiving antennas M 2
Number of information symbols in one frame 900
Error correction code Convolutional code (coding rate: 1/2, constraint length 3)
Modulation method BPSK
Decoder Log-Map decoder Error coding rate 1/2
Number of repetitions 5
F (b ′ _n (k)) = α × b ′ _n (k)
It was.

FIG. 28 shows the BER characteristics of the multi-output turbo receiver shown in FIG. 1 and the multi-input multi-output turbo receiver reflecting the error correction decoding result in b ′ _n (k). The latter is shown in white. The circle represents the first repeat, the downward triangle represents the second repeat, the diamond represents the third repeat, the left triangle represents the fourth repeat, and the right triangle represents the fifth repeat. FIG. 28A shows a simulation result of the BER characteristic with respect to Eb / No when α = 0.2, and FIG. 28B shows a simulation result of the BER characteristic with respect to α when Eb / No = 6 dB. Here, α = 0 is equal to b ′ _n (k) = 0. From this Figure 28A, b 'in the _n multiple-input multiple-output receivers that reflects the error correction decoding result (k), as compared to the multiple input multiple output turbo receiver shown in FIG. 1, the number of repetitions of the third and subsequent In the case, the improvement effect is large with respect to the BER at the time of the previous iterative decoding, and when the number of repetitions is 3 or more, when compared with the required E _b / N ₀ that achieves each BER in the range of BER> 10 ⁻⁴ The multi-input multi-output turbo receiver in which the error correction decoding result is reflected in b ′ _n (k) has a gain of about 0.5 dB or more compared to the multi-input multi-output turbo receiver shown in FIG. Further, in the fifth repetition of Eb / No = 6 dB, BER = 10 ⁻⁵ BER was achieved, and it can be seen that the BER could be reduced to 1/10 or less compared to that shown in FIG. From FIG. 28B, improvement is obtained in the range of 0 <α <0.6 as the value of α. On the contrary, when α is larger than 0.6, the BER characteristic deteriorates, and a correct decoding result is obtained. It becomes impossible. From this result, it is understood that the optimum value of α in this case is 0.2. However, the value of α is not limited to the optimum value, and the appropriate range of α having an improvement effect may be changed depending on the number of users to receive, the propagation environment including interference, the number of antennas to receive, and the like. Also, the value of the optimum value α may take other values.

In the case of BPSK modulation, the number of users (transmitters) is N, the number of multipaths of each transmitter is Q, the number of antennas of the receiver is M, and in the case of BPSK modulation, the conventional single-user turbo receiver has multiple outputs ( The amount of calculation in the equalizer when expanded to (MIMO) is on the order of 2 ^{N (Q-1)} as described above, but according to the turbo reception method of the third invention, the order of N (MQ) ³ Just do it. For example, if N = 8, Q = 20, and M = 8, 2 ^{N (Q-1)} ≈5 × 10 ⁴⁵ , but N (MQ) ³ ≈37 · 10 ⁷ , which is the turbo receiving method of the second invention. Therefore, the calculation amount can be significantly reduced.

According to the turbo receiving method of the second aspect of the present invention, it was confirmed by performing a simulation under the following conditions that a good bit error rate characteristic can be obtained. The channel matrix H is assumed to be known.
Number of users N 4
Number of multipaths Q 5 for each user
Number of receiving antennas M 2
Number of information symbols in one frame 900 bits Error correction code rate 1/2, constraint length 3 convolutional code Doppler frequency 1000 Hz (Rayleigh fading)
Modulation method BPSK
Transmission speed 20Mbps
Decoder Log-MAP decoder Number of repetitions 6 times Channel estimation is ideal FIG. 29 shows a simulation result of this BER (bit error rate) characteristic. The horizontal axis is the average E _b (bit energy) / N _o (noise power), f _d is the Doppler frequency, and T _s is the transmission symbol period. The MRC shown in this graph is a BER characteristic obtained when Viterbi decoding is performed on a signal after maximum ratio combining (MRC) in an order 10 (2 antenna × 5 paths) diversity channel. Corresponds to the BER characteristics when the interference is completely canceled. That is, the quality of the receiver can be evaluated by how close the BER after repetition is to the MRC curve. As shown in FIG. 27, according to the turbo reception method of the second invention, the BER decreases as E _b / N _o increases, and the BER characteristic approaches the BER characteristic of MRC as the number of repetitions increases. It turns out that it is very close to MRC. That is, it was confirmed that the multi-output turbo receiver according to the turbo receiving method of the third invention operates properly even under severe conditions of 4 users, 5 paths each, and 2 receiving antennas.

The figure which shows the function structure of the system containing the Example of the turbo receiver of this 1st invention. FIG. 2 is a diagram illustrating a specific functional configuration example of a multi-output equalizer 31 in FIG. 1. The flowchart which shows the Example of the turbo receiving method of this 1st invention. A is a figure which shows the example of a frame structure, B is a figure which shows the process in each repetition for demonstrating the repetition channel estimation method in 4th invention. The figure which shows the function structural example for taking out the reliable hard decision symbol. The flowchart which shows the example of the process sequence of the repetition channel estimation in this invention. A is a diagram showing an example of a functional configuration of a part of the equalizer 31 in the second invention that reflects an error correction decoding result of a signal to be detected, and B is a diagram showing an example of the processing procedure. The figure which shows the example of the receiver which performs a turbo equalizer repeatedly. The figure which shows the example of the receiver which repeats RAKE reception-turbo decoding. The figure which shows the example of the receiver which repeats adaptive array antenna reception-turbo decoding. The figure which shows the outline of a turbo equalizer and a turbo decoder. The figure which shows the outline of the receiver which repeats the process which uses an estimation channel with respect to a received signal, and the decoding process of the processed signal. The flowchart which shows the example of the schematic process sequence of the receiving method which repeats the process which uses an estimation channel with respect to a received signal, and the decoding process of the processed signal. A is a diagram illustrating an example of a frame configuration, and B is a diagram illustrating an iterative process of estimating a channel H and a noise covariance matrix U when a received signal includes noise other than white Gaussian noise. The figure which shows the one part functional structure example of the equalizer using estimation of the noise covariance matrix U. The flowchart which shows the example of the process sequence which repeats the channel value estimation using estimation of the noise covariance matrix U, and a decoding process. The figure which shows the principle of the turbo receiver by this 3rd invention. The figure which shows the function structural example of the turbo receiver by this 3rd invention. FIG. 19 is a diagram showing a specific example of a functional configuration of a multiuser (previous stage) equalizer 71 in FIG. 18. The flowchart which shows the example of the process sequence of the turbo reception method by this 3rd invention. The figure which shows the other functional structural example of the multistage equalization part in 3rd invention. The figure which shows the system structural example to which the Example of 1st invention (2) was applied. Error rate characteristic diagram of a turbo receiver to which the first invention (1) is applied (assuming that the channel is completely estimated, E _b (bit energy): N _o for two users is noise energy). The figure which shows the error rate characteristic at the time of performing channel estimation repeatedly by changing a threshold value (Th). In the 4th invention, especially the error rate characteristic figure of the turbo receiver which uses repetitive channel estimation. The figure which shows the error rate characteristic of the turbo receiver which uses estimation of the noise covariance matrix U. The figure which shows the error rate characteristic of the turbo receiver shown in FIG. The figure which shows the error rate characteristic of the Example of the 2nd invention which reflected the error correction decoding result of the signal to detect. The figure which shows the simulation result of the error rate characteristic of the turbo receiver of this 3rd invention. The figure which shows the concept of a MIMO system. The figure which shows the function structure of the conventional turbo transceiver for single users.

Claims (8)

The channel value as the channel characteristic of the received signal is estimated from the received signal and the known signal as the reference signal, the received signal is processed using the estimated channel value, and the processed signal is decoded. In the receiving method of repeatedly performing the process using the estimated channel value and the decoding process for the same received signal,
The certainty of the decoded hard decision information symbol is determined from the value of the soft decision information symbol, and a hard decision information symbol having a certainty or more of a certain value is also used as a reference signal for the next channel estimation. Turbo reception method. When the received signal is processed using the estimated channel value, as a covariance matrix U of noise components in the received vector y (k) as the received signal , σ ² I (σ ² is the variance of the Gaussian distribution) value, I is turbo reception method according to claim 1, wherein the benzalkonium to calculate the identity matrix). When the processing using the channel values the estimated on the received signal, the covariance matrix U of noise components of the received signal vector in the y (k) as the received signal (the estimated value and U ^.) using the estimated channel matrix H ^ received signal vector y a (k),
b ₁ (k + q) to b _N (k + q) are the reference signal composed of the known signal and the hard decision information symbol having the certainty of the predetermined value or more, and Tr is the reference signal length
U ^ = Σk _{= 0} ^Tr (y (k) -H ^ B (k)) * (y (k) -H ^ B (k)) ^{H
B (k) = [b T} (k + Q-1) ... b T (k) ... b T (k-Q + 1)] T
b (k + q) = [b ₁ (k + q)... b _N (k + q)] ^T (q = −Q + 1... Q−1)
The turbo reception method according to claim 1, wherein the estimation is performed by calculating.   In the repetition of the process using the estimated channel value and the decoding process, a linear equalization filter is determined based on the estimated channel value, the received signal is processed by the linear equalization filter, and the processed signal is decoded. The turbo reception method according to claim 1, wherein the turbo reception method is a repetition of the above.   The repetition of the process using the estimated channel value and the decoding process is performed by performing a rake combining process for compensating for the phase rotation received by each symbol in the transmission path in the rake combining processing unit by the estimated channel value. The turbo reception method according to claim 1, wherein the signal is subjected to repetition of decoding by a turbo decoder the signal subjected to the rake synthesis process.   In the repetition of the process using the estimated channel value and the decoding process, the weight for determining the antenna directivity is set for the adaptive array antenna receiving unit based on the estimated channel value, and the output of the adaptive array antenna receiving unit is set. The turbo reception method according to claim 1, wherein the turbo reception method is repetition of decoding by a turbo decoder.   The output of the adaptive array antenna receiving unit is subjected to a rake combining process for compensating the phase rotation received by each symbol in the transmission path by the estimated channel value in the rake combining processing unit, and the signal subjected to the rake combining process is converted to the turbo signal. The turbo reception method according to claim 6, wherein the turbo reception method is supplied to a decoder. The channel value, which is the channel characteristic of the received signal, is estimated from the received signal and the known signal as the reference signal, the received signal is processed using the estimated channel value, and the processed signal is decoded. In a receiver that repeatedly performs processing using the estimated channel value and decoding processing on the same received signal,
Means for determining whether the probability of the decoded hard decision information symbol is equal to or greater than a predetermined value, based on whether the value of the soft decision information symbol is equal to or greater than a threshold, and hard decision information determined to be probable A turbo receiver comprising: a previous symbol storage unit in which stored content is updated and stored using symbols, wherein the stored content of the previous symbol storage unit is used as a reference signal for the next channel estimation.

Images