JP3263389B2 - Communication path decoding method and apparatus - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a communication channel decoding method and apparatus for protecting digitized voice information from transmission errors in a digital mobile communication system for public business or the like.

[0002]

2. Description of the Related Art When transmitting a voice signal in mobile communication, a high-efficiency voice coding technique is used in order to improve the frequency use efficiency. This high-efficiency speech coding scheme is a waveform coding scheme that performs efficient compression by actively utilizing the features of speech, such as adaptive differential pulse code modulation (ADPCM), A method such as CELP, VSELP, PSI-CELP, which divides the frame into frames at intervals and transmits the speech synthesis parameters and the residual excitation signal, and transmits only the speech synthesis parameters to obtain a lower transmission rate LPC vocoder, IMBE,
Alternatively, there is a method proposed by the present applicant in Japanese Patent Application No. 11-223804.

In order to protect speech information bits encoded by such a high-efficiency speech encoding technique from transmission errors, CRC (Cyclic) is applied to parameters that are important for hearing.
A method of performing maximum likelihood decoding (Viterbi decoding) by soft-decision decoding is widely used on the receiving side to decode and transmit speech information bits by applying error detection coding and convolutional coding by Redundancy Check. I have. Soft decision decoding is
When a digital signal is transmitted on a transmission line where a noise source is present, the digital signal is estimated by determining a noise-added signal as a multi-valued signal, and the transmitted signal is regarded as "0" or "1". The error rate can be improved as compared with hard-decision decoding in which a binary signal of "" is determined. References regarding soft decision decoding include [1] Kiyoshi Arai, “Introduction to Turbo Codes”, WS No.195, PP.189-197, Trikeps Co., Ltd. (1
999.10). For maximum likelihood decoding (Viterbi decoding), see [2] Toru Inoue's "Combat Error Correction Techniques", Trikeps Monographs (TR) 5, PP.143-166.

FIG. 9 is a diagram showing a baseband processing unit of a conventional speech communication system using soft decision decoding.
(A) shows the configuration on the transmitting side, and (b) shows the configuration on the receiving side.
On the transmitting side of (a), an input audio signal (a1) is coded by an audio encoder (11), and an audio information bit (b1)
Is output. As this speech encoder (11), Japanese Patent Application No. Hei.
In the case of using a speech coding method in which the coding rate proposed in 1-223804 is 1.6 kbps, the speech information bit (b1)
In LSF (Line Spectrum Fr)
equencies) Coefficients (spectral envelope information), gain information, periodic / aperiodic pitch / voiced / unvoiced information codes, and high-frequency voiced / unvoiced flags are included. In the following description, for simplicity, audio information bits per audio encoded frame are considered as a data transmission unit (or transmission frame).
The subsequent processing is executed for each data transmission unit (transmission frame).

[0005] The error detection encoder (12) performs error detection encoding by CRC on the audio information bits (b1) that are important for perception, and converts the bit sequence (c1) with the CRC code added thereto. Output. The error correction encoder (13) performs, for example, convolutional coding on bits that are perceptually important in the bit sequence (c1), and performs bit correction (d
Output 1). Here, error protection is not applied to bits that are insignificant to the auditory sense in the bit string (c1), and are not processed and output as part of the bit string (d1) after error correction coding. Note that the error detection coding and the error correction coding may be performed on all of the audio information bits (b1). Next, the interleaver (14) interleaves the bit string (d1) after error correction coding as a measure against burst errors, and outputs the resulting interleaved bit string (e1).
The modulator (15) digitally modulates the interleaved bit string (e1) using, for example, π / 4-QPSK, and the resulting modulated wave (f1) is transmitted through a radio unit (not shown) and a transmission antenna through a communication path. Is output to

On the receiving side (b), a modulated wave (g1) received through a receiving antenna and added with noise in a communication channel
Is demodulated by a demodulator (16) by, for example, a delay detection method. The N-bit A / D converter (17) quantizes the analog signal (h1) as a demodulation result with N (for example, N = 3 to 8) bits, and outputs soft-decision demodulated data (i1). This quantization is performed as shown in FIG. FIG. 4 shows a case where quantization is performed with N = 3 for simplicity. In the figure, zj (j =
1,..., 8) denote soft-decision demodulated data as quantization results. In the figure, the horizontal axis represents the level of the analog output signal (h1) from the demodulator (16), and A and -A respectively correspond to the received signal amplitude at the time of noiseless, and the coded data 0,
Corresponds to 1. T is a quantization interval.

The deinterleaver (18) deinterleaves the soft-decision demodulated data (i1) and outputs soft-decision demodulated data (j1) as a result. The Viterbi decoder (19) receives the soft-decision demodulated data (j1), performs soft-decision decoding, and decodes speech information bits and a CRC code (k1). Table 1 shows examples of soft decision decoding metrics used here.

[Table 1]

The error detector (20) separates the speech information bits and the CRC code (k1) into a CRC code and speech information bits, and uses the CRC code to perform error detection processing on the decoded speech information bits. I do. As a result, a decoded speech information bit (l1) and an error detection flag (m1) indicating the presence or absence of a transmission error are output. If the error detection flag (m1) indicates that there is a transmission error, the voice decoder (21) replaces all parameters of the current frame with normal parameters of the previous frame before performing voice decoding (parameter interpolation). Then, it performs audio decoding processing and outputs a reproduced audio signal (n1).

In the conventional system described above, the A / D
In the converter (17), a quantization interval T for generating soft-decision demodulated data is fixedly set in advance. Decoder characteristics are greatly affected by the quantization interval T. If this interval is too small, the determination of each level becomes unstable. It is known that the optimal value T _opt of T differs depending on the noise situation and is approximated by the following equation (reference [2] above). _{T opt = α [N 0/} 2] 1/2 (1) where, alpha is 0.6 to 0.7, N ₀ is the noise power. The value of the quantization interval T used in an actual system is designed so as to cover the dynamic range of the received signal level, cover the fluctuation range, and maximize the soft decision gain.

However, fading occurs on a propagation path in a real environment in which a digital mobile communication system such as an automobile / mobile phone system is used, and the received signal level fluctuates greatly. Accordingly, the fluctuation width greatly deviates from the fluctuation width assumed when the quantization interval T is determined as described above. Therefore, the set quantization interval T is not an appropriate value, and
This results in a decrease in the soft decision gain and an increase in the error rate. To solve this problem, for example, as proposed in JP-A-6-224959, an average level of a received signal is obtained for each transmission unit, and a quantization range (interval) is determined by referring to a table from the average level. ) A quantization range and level adjuster for determining a constant and generating soft-decision demodulated data based on the constant. Soft-decision is performed by adjusting so that the smaller the average level, the narrower the quantization range and the smaller the level. A method of obtaining demodulated data has been proposed.

[0011]

In the above-mentioned conventional method, an average level of a received signal is obtained for each transmission unit, and a quantization range and a level for producing soft-decision demodulated data are adjusted from the average level. are doing. Therefore, when the level fluctuation due to fading is relatively gradual with respect to the time width of the data transmission unit (the time width of the slot when the access method is TDMA), that is, the level fluctuation within the time width of the transmission unit is relatively large. It is effective in the case where it is moderate. However, when the maximum Doppler frequency is high or the time width of the transmission unit is large, the fluctuation of the received signal level within the time width of the transmission unit is rapid and sharp, so that an appropriate quantization range is not obtained. The setting is not limited, and the obtained soft-decision demodulated data may not be an appropriate value, and the soft-decision gain may be reduced. In such a case, in the speech communication system shown in FIG. 9, the probability that the error detector (20) determines that there is an error (referred to as “error detection rate”) increases, and the speech decoder (21) The frequency with which the parameter interpolation is performed increases, and the clarity or intelligibility of the reproduced sound decreases.

Therefore, the present invention reduces the error detection rate even if the received signal level fluctuates rapidly within the time width of the transmission unit, that is, by reducing the frequency of parameter interpolation in the speech decoder. It is another object of the present invention to provide a communication channel decoding method and apparatus capable of suppressing a decrease in clarity or intelligibility of reproduced voice quality.

[0013]

In order to achieve the above object, a communication channel decoding method according to the present invention comprises the steps of: performing error detection coding on all or a part of speech information bits and then performing error correction coding; Channel decoding method for data channel-encoded by using a soft decision demodulation data which is a result of quantizing an analog signal obtained by demodulating a received signal at a predetermined quantization interval. A soft decision decoding step for decision decoding, an error detection step for performing error detection processing on the soft decision decoding output, and a step of quantizing the analog signal when it is determined that there is an error in the error detection step. Change the quantization interval to another predetermined set value and create the soft decision demodulated data again,
On the other hand, it has an iterative decoding step of repeating the soft decision decoding step and the error detection step until it is determined that there is no error in the error detection step, or until a predetermined number of repetitions is reached.

When it is determined that there is no error as a result of the processing in the iterative decoding step, a feature parameter is reproduced from the decoded speech information bit string, and an error is overlooked by utilizing the statistical property of the feature parameter. Error detection step for determining whether or not an error has occurred, and a step of correcting the error detection result to an error when it is determined in the error oversight step that an error has been detected. It is. further,
The error oversight detection step uses gain information as the feature parameter to determine an absolute value of a difference value between the gain of the past frame and the gain of the current frame, and sets the absolute value of the difference value to a predetermined threshold value. If greater,
This is to determine that an overlooked error has occurred. Still further, the error oversight detecting step uses the pitch cycle information as the feature parameter, obtains an absolute value of a difference value between a pitch cycle of a past frame and a pitch cycle of a current frame, and calculates an absolute value of the difference value. Is larger than a preset threshold value, it is determined that an error has been missed. Furthermore, the error oversight detecting step uses the spectrum envelope information as the feature parameter, obtains a Euclidean distance between the spectrum envelope information of the past frame and the spectrum envelope information of the current frame, and the Euclidean distance is set in advance. If it is larger than the threshold value, it is determined that an error is missed. Furthermore, in the error oversight detection step, a combination of gain information, pitch period information, and spectrum envelope information is used as an error oversight detection condition.

Still another channel decoding method according to the present invention is to perform error detection coding on all or a part of audio information bits, and then perform error correction coding on the data to perform channel coding data. Communication channel decoding method,
A soft-decision decoding step of performing soft-decision decoding using soft-decision demodulated data that is a result of quantizing an analog signal obtained by demodulating a received signal at a predetermined quantization interval, and performing error detection processing on the soft-decision decoded output In the error detection step, when it is determined that there is an error in the error detection step, the number of quantization bits when quantizing the analog signal is set to a larger value, and the soft decision demodulated data is created again. In contrast, the method further comprises an iterative decoding step of repeating the soft decision decoding step and the error detection step until it is determined that there is no error in the error detection step or until a predetermined number of repetitions is reached.

Still further, the communication path decoding apparatus of the present invention comprises:
A channel decoding apparatus for channel-coded data by performing error detection coding on all or some audio information bits, and then performing error correction coding, and demodulates a received signal. A Viterbi decoder that performs soft-decision decoding using soft-decision demodulated data that is the result of quantizing an analog signal at a predetermined quantization interval, and an error detector that performs an error detection process on an output of the Viterbi decoder. In response to the output indicating that there is an error from the error detector, the quantization interval when quantizing the analog signal is changed to another predetermined set value to produce the soft-decision demodulated data again, And an iterative decoding controller for repeatedly operating the Viterbi decoder and the error detector until the error detector determines that there is no error, or until a predetermined number of repetitions is reached. It is. Furthermore, when it is determined that there is no error during the repetition operation, a feature parameter is reproduced from the decoded audio information bit string, and whether or not an error oversight using the statistical property of the feature parameter has occurred is determined. It has an error oversight detector that makes a determination and, when it is determined that an error oversight has occurred, corrects the error detection result to an error. Furthermore, the error oversight detector determines whether or not the error oversight has occurred by using any one of gain information, pitch period information or spectrum envelope information or a combination thereof as the feature parameter. It is.

[0017]

FIG. 1 is a block diagram showing a configuration of a channel decoding apparatus to which a channel decoding method of the present invention is applied. However, the processing on the transmitting side is the same as the conventional example shown in FIG. 9A. In FIG. 1, a modulated wave (a2) to which noise has been added in a communication path, received through a receiving antenna, is demodulated by a demodulator (30), for example, by a delay detection method. The M-bit A / D converter (31) quantizes the analog signal (b2) as a demodulation result by, for example, M = 15 bits, and outputs digitized demodulated data (c2). The quantization range here is set to an optimal value so as to cover the fluctuation range in consideration of the level fluctuation of the received signal due to fading or the like. The subsequent processing is performed for each data reception unit (corresponding to the data transmission unit (transmission frame)).

The deinterleaver (32) deinterleaves the demodulated data (c2) digitized by 16 bits and outputs the result (d2) to the buffer (33). The buffer (33) holds digitized demodulated data for a reception unit (one decoded frame). Level adjuster (34)
Is the scaling information (k2) from the iterative decoding controller (38)
, The level of demodulated data (e2) for one decoded frame, which is output from the buffer (33), is adjusted, and (f2) is output. The control of the iterative decoding will be described later.
An N-bit quantizer (35) requantizes the digitized demodulated data (f2) after the level adjustment with N bits (N <M, N is 3 to 8 bits), and quantizes N (for example, 3) bits. The soft decision demodulated data (g2) thus output is output. Here, the digitized demodulated data (f2) is M-bit data quantized by the M-bit A / D converter (31) as described above, and is an N-bit quantizer (35) Performs the requantization by extracting N bits at a predetermined position in the M bits. The Viterbi decoder (36) receives the soft-decision demodulated data (g2) quantized by 3 bits, performs soft-decision decoding, and decodes speech information bits and a CRC code (h2). The metric of soft decision decoding used here uses the one shown in Table 1 above.

The error detector (37) separates the speech information bits and the CRC code (h2) into a CRC code and speech information bits, and uses the CRC code to perform error detection processing on the decoded speech information bits. I do. As a result, a decoded audio information bit (i2) and an error detection flag (j2) indicating the presence or absence of a transmission error are output. The audio decoder (39) performs audio decoding processing on the audio information bit (i2), and outputs a reproduced audio signal (l
Output 2). However, if the error detection flag (j2) indicates that there is a transmission error, before performing the audio decoding process, after replacing all parameters of the current frame with normal parameters of the previous frame (parameter interpolation), the audio decoding process is performed. Then, a reproduced audio signal (l2) is output.

Hereinafter, the iterative decoding process will be described. The iterative decoding controller (38) inputs the error detection flag (j2) output from the error detector (37), and when it indicates that there is an error, demodulated data (e2) quantized by M bits. N
The quantization interval at the time of requantization with (<M) bits is changed to another predetermined set value, and the soft decision demodulated data (g
2) Make. This principle will be described with reference to FIGS. FIG. 4 is an explanatory diagram showing a normal 3-bit quantization method, and zj (j = 1,..., 8) indicates soft decision demodulation data as a quantization result. In the figure, the horizontal axis indicates the level of the digitized demodulated data (e2), and A and -A respectively correspond to the received signal level in the absence of noise, and correspond to the coded data 0 and 1, respectively. T is a quantization interval.

The channel decoding method proposed in the present invention is shown in FIG.
Judgment demodulation data (g2) created by normal quantization shown in
If the error detection result (error detection flag (j2)) after the soft decision Viterbi decoding indicates that there is an error, the quantization interval T is multiplied by x as shown in FIG. Generate decision demodulation data (g2). FIG. 3A shows x = 2,
(B) shows a case where x = 0.5. Here, the quantization interval T
Is obtained by scaling the demodulated data (e2) quantized by M bits by 1 / x times,
Is equivalent to performing the normal quantization shown in FIG. Therefore, in order to realize this processing, a predetermined 1 / x set value (scaling information (k2)) is output from the iterative decoding controller (38) to the level adjuster (34), After scaling the demodulated data (e2) digitized by (34) by 1 / x, the normal quantization shown in FIG. 4 may be performed.

The iterative decoding controller (38) outputs an error detection result (error detection flag (j2)) after soft-decision Viterbi decoding to the soft-decision demodulated data (g2) regenerated in this manner. Or until the number of repetitions reaches a preset number of times, the quantization interval T is changed to another set value, soft-decision demodulated data is created again, and Viterbi decoding and error detection processing are repeated. The preset value in the iterative decoding controller (38) is, for example, the number of repetitions R = 10, x
Is set to the lower limit value X1 = 0.1 and the upper limit value X2 = 2.0, and at the time of the r-th repetition, x = X1 + (r-1) (X2-X1) / (R-1), r = 1,2 , ..., R (2) is set as x, and 1 / x is output as scaling information (k2). As a result, the quantization interval T becomes T × 0.1 to T
Iterative decoding processing is performed using the soft-decision demodulated data in the range of × 2.0. Note that the range of x is not limited to the above example, and the number of repetitions R can be set arbitrarily.

The decoding processing operation according to the first embodiment of the present invention will be described with reference to the flowchart of FIG. First, in step S51, received data (corresponding to (c2) in FIG. 1) for one frame (transmission unit) is fetched, and deinterleaving is executed in step S52. Next, in step S53, the number of repetitions R of the iterative decoding, the lower limit value X1 and the upper limit value X2 of the multiple x with respect to the quantization interval T are set, and the repetition index r is set to 0 in step S54.
Set to. Hereinafter, a repetition loop (step
Enter S55-S65).

First, it is determined in step S55 whether or not r = 0. If r = 0, the flow branches to step S56. Otherwise, the flow branches to step S56. Initially, in step S54, r =
Since it is set to 0, the flow branches to step S56, and x =
1.0. This corresponds to the execution of the normal quantization processing shown in FIG. In the case of iterative decoding (r ≠ 0), the value of x according to the above equation (2) is set in step S57. Then, the process proceeds to step S58 to multiply the received data by 1 / x. This corresponds to the scaling processing by the level adjuster (34). Next, proceeding to step S59, N-bit quantization is performed (N-bit quantizer (35)). Then, in step S60, soft decision Viterbi decoding is performed (soft decision Viterbi decoder (36)), and r
If = 0, the decoded data (h2) is stored (step S61).

Then, the flow advances to step S62 to perform error detection using the soft decision decoding result (error detector (37)).
As a result, if there is no error, the error detection flag crc_err = 0 is set, and if there is an error, crc_err = 1 is set. And c
If rc_err = 0 (no error), the decoding process is terminated as it is, and if crc_err = 1 (error is present), the process proceeds to step S64 (step S63). In step S64, 1 is added to the repetition index r. Then, step S65
Then, it is determined whether or not r exceeds the predetermined number of repetitions R. If r ≦ R, the process returns to the step S55, and x, that is, the scaling information (k2) is changed to generate soft-decision demodulated data again. The soft decision decoding and error detection processing are performed. On the other hand, when r> R, that is, when the error detection result has an error (crc_err = 1) even after R times of iterative decoding, the process proceeds to step S66, and the normal quantization processing stored in step S61 is performed. The decoded data at that time is replaced by the decoded data (h2) based on the result. In this case, as described above, parameter interpolation is performed in the audio decoder (39).

According to the first embodiment of the present invention, iterative decoding processing is performed while changing the quantization interval T until it is determined that there is no error or until a predetermined number of repetitions is reached. Is performed, the error detection rate of the decoding result is significantly reduced, and the number of frames in which parameter interpolation is performed in the speech decoder (39) can be reduced. Here, the error detection rate is a ratio of frames in which an error detection result indicates an error in all frames. Therefore, even if the received signal level fluctuates rapidly within the time width of the data transmission unit, it is possible to suppress a decrease in the clarity or intelligibility of the reproduced sound.

In the above-described first embodiment, decoding is repeated until no error is detected by iterative decoding. Therefore, depending on the capability of the error detection code (CRC code), the error overlooking rate may be significantly increased. There is. Here, the error oversight is a decoding error in which it is determined that there is no error in the error detection result even though a code error has occurred. The error oversight rate is a ratio of frames in which error oversight occurs in all frames. In a frame where an error has been missed, even if there is an error in the decoded feature parameters, parameter interpolation is not performed, and speech decoding processing is performed using those parameters as they are. , The sound quality may be degraded.

A description will now be given of a second embodiment of a communication channel decoding method and apparatus according to the present invention which can solve such a problem. In the second embodiment,
In order to solve the above-mentioned problem, by using the statistical properties of the speech feature parameters, by detecting missed errors (errors that cannot be detected even though an error has occurred on the communication channel), the error missed rate is reduced, The sound quality of the reproduced sound is further improved. FIG. 2 is a block diagram showing a configuration of the second exemplary embodiment of the present invention. As is clear from a comparison between FIG. 2 and FIG. 1, the second embodiment is different from the first embodiment in that the error detector (37) and the speech decoder (39) In addition, an error oversight detector (40) is added between them, and other same components are denoted by the same reference numerals as those in FIG.

As described above, the speech encoder on the transmitting side ((11) in FIG. 9) employs a high-efficiency speech encoding system, and the speech information bits output therefrom include gain information and pitch period. Various characteristic parameters such as information and spectrum envelope information are included. Therefore, the error overlook detector
(40) reproduces a characteristic parameter from the decoded audio information bit sequence (i2) when it is determined that there is no error as a result of the iterative decoding (excluding the first normal decoding), Utilizing the statistical properties of (1), an overlooked error is detected.If the overlooked error is detected, the error detection flag (j2) is corrected to include an error, and the decoded speech information bit sequence or the reproduced feature parameter ( a3) and the error detection flag (b3) are output to the audio decoder (39). As a result, when an error is missed, the error detection flag can be corrected to include an error, and deterioration of the sound quality of the reproduced sound can be prevented.

Hereinafter, a specific method for detecting an overlooked error in the overlooked error detector (40) will be described. One method uses the gain information as the above-mentioned feature parameter, finds the absolute value Gd of the difference between the gain Gp of the previous frame and the gain G of the current frame as shown in the following equation (3), and When the absolute value Gd of the difference is larger than the threshold Thg, it is determined that an error is missed, and the error detection result is corrected to an error. Gd = | G−Gp |> Thg (3) Alternatively, using the pitch period information as the feature parameter, the pitch period Pp of the previous frame is calculated as shown in equation (4).
And the absolute value Pd of the difference from the pitch period P of the current frame is obtained and compared with a preset threshold value Thp. If the absolute value Pd of the difference is larger than the threshold value Thp, an error is overlooked. Is determined, and the error detection result is corrected to include an error. Pd = | P-Pp |> Thp (4)

Alternatively, spectrum envelope information (eg, LSP (Line Spectrum Pair)) may be used as the feature parameter.
LS of the previous frame as shown in equation (5).
The Euclidean distance between the P coefficient fpi (i = 1,2, ..., p: p is the analysis order) and the LSP coefficient fi (i = 1,2, ..., p) of the current frame
fd is obtained and compared with a preset threshold Thf. If the Euclidean distance fd is larger than the threshold Thf, it is determined that an error is missed, and the error detection result is corrected to an error.

(Equation 1)

Alternatively, a combination of the above-described gain information, pitch cycle information, and spectrum envelope information is used as a condition as a characteristic parameter used in the detection of the missed error. For example, if the expression (3) or the expression (4) is satisfied, it is determined that an error is overlooked, and the error detection result is corrected to an error. The condition for detecting an overlooked error according to the above equations (3), (4) and (5) is that a temporal change of characteristic parameters (gain, pitch period, spectrum envelope information, etc.) in a voiced section of a voice is usually moderate. This is based on the statistical property that the possibility of a sudden change is large in a frame in which an error has occurred.

The decoding operation according to the second embodiment of the present invention will be described with reference to the flowchart of FIG. In the figure, steps S67 to S76 correspond to the operation of the error oversight detector (40). The other steps are the same as those in the first embodiment shown in FIG. However, in this example, the equation (3) using the gain information and the pitch cycle information are used (4)
The combination of equations is used as a condition for detecting missed errors. When the error detection flag becomes no error (crc_err = 0), the operation of the error oversight detector (40) is started,
First, in step S67, gain information G and pitch period information P are decoded from the decoded data. If this is the first received frame, the process jumps to step S76 (step S68). Otherwise, the process proceeds to step S69, where it is determined whether 0 <r ≦ R.
When 0 <r ≦ R, that is, when iterative decoding
When c_err = 0, the steps following step S70 are executed, otherwise, jump to step S76.

In step S70, the absolute value Gd of the difference between the gain Gp of the previous frame and the gain G of the current frame Gd = |
Calculate G-Gp |. Next, in step S71, Gd is compared with a preset threshold value Thg, and if Gd> Thg, it is determined that an error has been overlooked, and the process proceeds to step S72, where crc_err
= 1 (with error), and the decoding process ends. G
When d ≦ Thg, the process proceeds to step S73, and the absolute value Pd = | P−Pp | of the difference between the pitch period Pp of the previous frame and the pitch period P of the current frame is calculated. And step
In S74, Pd is compared with a preset threshold Thp, and Pd> Th
If it is p, it is determined that an error was overlooked, and step S75
To set crc_err = 1 and end the decoding process.
If Pd ≦ Thp, it is determined that no error was overlooked, and the process proceeds to step S76, where the gain G and pitch period P of this frame are stored as the gain Gp of the previous frame and the pitch period Pp of the previous frame, and the decoding process is performed. finish.

According to the second embodiment of the present invention, an overlooked error is detected by using the statistical properties of the characteristic parameters, and the error detection flag is corrected. As compared with the case of the above embodiment, it is possible to prevent the generation of an impulsive sound or the like due to the miss of the error, and it is possible to further improve the reproduced sound quality.

FIG. 8 is a diagram showing the measurement results of the error rate (BER) vs. error detection rate and error oversight rate characteristics when the second embodiment is applied. Here, in the above-described iterative decoding control, R = 10, X1 = 0.1, X2 = 2.0, and the threshold value Thg of the gain of Expression (3) in the error oversight detection.
= 10.0 dB, and the threshold value Thp of the pitch period in Expression (4) = 15 samples. FIG. 11 shows the result when the number of quantization bits N = 3 in the N-bit quantizer (35), but the same effect is obtained when the measurement is performed with N = 8 and 15. Table 2 shows the specifications of the channel model used for this measurement.

[0037]

[Table 2]

In FIG. 8, BER is an error rate measured at the output of the demodulator (30) (by hard decision demodulation). In the figure, the solid line marked as not applied is the result of the conventional example shown in FIG. 9, that is, the result of ordinary decoding, and the broken line marked as applied is the decoding result of the second embodiment described above. It is. (A) is the maximum Doppler frequency fD = 10
Hz, delay time Delay = 20μs, (b) fD = 20Hz, Delay
= 20 μs, (c) shows the measurement results for fD = 40 Hz and Delay = 20 μs. From this result, it can be seen that by using the second embodiment, the error detection rate decreases as the BER increases. Also, the larger the maximum Doppler frequency, the greater the reduction rate when the BER is high. Although the error oversight rate is slightly increased, the reduction of the error detection rate is larger and does not pose a problem.

In each of the embodiments described above, the level adjuster (34) is used, and at the time of iterative decoding, the demodulated data (e2) is scaled by the scaling information (k2) from the iterative decoding controller (38). The quantization interval T
Had changed. However, the method of changing the quantization interval T is not limited to this, and the quantization interval T may be changed by changing the number of bits N of the N-bit quantizer (35). it can. In general, when performing soft decision decoding, it is known that the error characteristics are improved as the number of quantization bits is increased (however, if the number of quantization bits is 8 or more, the error characteristics are saturated and the error characteristics are not improved). Therefore, when performing the above-mentioned iterative decoding, the number of quantization bits in the N-bit quantizer (35) is increased to generate the soft decision demodulation data again, and the soft decision decoding processing and the error detection processing are performed. A description will be given of a third embodiment of the present invention which is repeated. In this case, the N
Although only the number N of quantization bits of the bit quantizer (35) may be changed, here, the level adjuster (34) employed in the first and second embodiments is used. An embodiment employing a combination of a method of scaling demodulated data and a method of changing the number N of quantization bits of the N-bit quantizer (35) will be described.

FIG. 3 is a block diagram showing a configuration of a third embodiment of the present invention in which the second embodiment is combined with a method of changing the number N of quantization bits of an N-bit quantizer (35). FIG. In this figure, the same components as those of the second embodiment shown in FIG. 2 are denoted by the same reference numerals, and the description will not be repeated. As is apparent from a comparison between FIG. 3 and FIG. 2, the iterative decoding controller (41) in this embodiment outputs scaling information (k2) to the level adjuster (34) and outputs N-bit data. Quantizer (3
The difference is that a quantization bit number indication signal (d3) is output to 5) and the Viterbi decoder (36). The N-bit quantizer (35) outputs the quantization bit number indication signal.
It is configured so that the value of N can be changed by the instruction of (d3), and the Viterbi decoder (36) can respond to the change in the number of quantization bits of the soft decision demodulated signal to be input. Is configured. Here, the N-bit quantizer (35) extracts and outputs N-bit data at a predetermined position in the demodulated data (e2) quantized with M bits as described above. And the number of bits to be extracted can be set by the quantization bit number indication signal (d3). Further, the Viterbi decoder (36), for example, each provided with a Viterbi decoder for soft decision demodulation data corresponding to each quantization bit number set by the quantization bit number indication signal (d3), What is necessary is just to select and use the soft-decision demodulation data of the corresponding bit number according to the quantization bit number instruction signal (d3).

The operation of the communication channel decoding method and apparatus of the present invention thus configured is such that the first normal decoding is performed with the quantization bit number N set to, for example, 3, and an error is detected in subsequent error detection. If it is determined that there is an error, the number of quantization bits when quantizing the demodulated data (e2) quantized by M bits is larger than N (for example, N
= 8), the soft-decision demodulated data is generated again, and the soft-decision decoding and the error detection are performed on the soft-decision demodulated data. Then, when the error is detected again, the scaling information (k2) is obtained in the same manner as in the second embodiment.
And performs the above-described scaling to generate the soft-decision demodulated data, and performs the soft-decision decoding and the error detection on the demodulated data. Hereinafter, similarly, the decoding process is repeated until no error is detected in the error detection (until it is determined that there is no error) or until a predetermined number of repetitions is reached. When the decoding process of the current frame is completed, a quantization bit number indication signal (d3) is output,
The number of quantization bits N is returned to 3. According to the third embodiment of the present invention, the number of repetitions R can be set smaller to obtain the same effect as that of the second embodiment.

[0042]

As described above, according to the channel decoding method and apparatus of the present invention, even if the received signal level fluctuates rapidly within the time width of the data transmission unit, the soft-decision demodulated data can be obtained. The error detection rate, that is, the frequency of parameter interpolation in the speech decoder, can be reduced by changing the quantization interval for generating a plurality of times and repeating soft decision decoding, and the clarity or intelligibility of the reproduced speech quality can be reduced. Can be suppressed.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a channel decoding device to which a first embodiment of a channel decoding method according to the present invention is applied.

FIG. 2 is a block diagram showing a configuration of a channel decoding device to which a second embodiment of the channel decoding method of the present invention is applied.

FIG. 3 is a block diagram showing a configuration of a channel decoding device to which a third embodiment of the channel decoding method of the present invention is applied.

FIG. 4 is a diagram for explaining a method of quantizing an analog output signal from a demodulator.

FIG. 5 is a diagram for explaining a method of quantizing an analog output signal from a demodulator during iterative decoding.

FIG. 6 is a flowchart for explaining processing in the first embodiment of the channel decoding method of the present invention.

FIG. 7 is a flowchart for explaining processing in a second embodiment of the channel decoding method of the present invention.

FIG. 8 is a diagram showing BER vs. error detection rate and error overlook rate characteristics when the second embodiment of the channel decoding method of the present invention is applied.

FIG. 9 is a block diagram showing a configuration of a conventional voice communication system using soft decision decoding.

[Explanation of symbols]

 REFERENCE SIGNS LIST 30 demodulator 31 M-bit A / D converter 32 deinterleaver 33 buffer 34 level adjuster 35 N-bit quantizer 36 Viterbi decoder 37 error detector 38, 41 iterative decoding controller 39 speech decoder 40 error oversight detection vessel

Continuation of the front page (56) References Japanese Patent Publication No. 8-21965 (JP, B2) Low bit rate speech codec for commercial mobile communication using mixed excitation type predictive coding, IEICE Transactions D- {I}, Japan, The Institute of Electronics, Information and Communication Engineers, Vol. J84-D- ▲ II ，, No. 4, 629-640 Study on error resilience of low bit rate voice codec for commercial mobile communications, 2000 IEICE General Conference, Japan, IEICE, 173

Claims (10)

(57) [Claims] 1. A communication channel decoding method for channel-coded data by performing error detection coding on all or a part of audio information bits, and then performing error correction coding. A soft-decision decoding step of soft-decision decoding using soft-decision demodulation data which is a result of quantizing an analog signal obtained by demodulating at a predetermined quantization interval, and performing error detection processing on the soft-decision decoded output. An error detection step, and in the error detection step, when it is determined that there is an error, a quantization interval for quantizing the analog signal is changed to another predetermined set value, and the soft decision is performed again. The demodulated data is generated, and the soft decision decoding step and the error detection step are performed on the demodulated data until the error detection step determines that there is no error or in advance. A communication channel decoding method comprising an iterative decoding step that repeats until a predetermined number of repetitions is reached. 2. As a result of the processing in the iterative decoding step, when it is determined that there is no error, a feature parameter is reproduced from the decoded audio information bit sequence, and an error is overlooked by using a statistical property of the feature parameter. An error oversight detection step of determining whether or not an error has occurred; and
2. The method according to claim 1, further comprising the step of correcting an error detection result to an error when it is determined in the error overlooking step that an error oversight has occurred.
The communication channel decoding method according to the above. 3. The error-missing detection step uses gain information as the feature parameter to determine an absolute value of a difference value between a gain of a past frame and a gain of a current frame, and calculates the absolute value of the difference value. 3. The communication channel decoding method according to claim 2, wherein when the threshold value is larger than a preset threshold value, it is determined that an error is missed. 4. The error-missing detecting step uses pitch cycle information as the feature parameter, finds an absolute value of a difference value between a pitch cycle of a past frame and a pitch cycle of a current frame, and calculates the absolute value of the difference value. 3. The communication channel decoding method according to claim 2, wherein when the absolute value is larger than a preset threshold value, it is determined that an error is missed. 5. The method according to claim 1, wherein the overlooking error detection step uses spectrum envelope information as the feature parameter to determine a Euclidean distance between the spectrum envelope information of the past frame and the spectrum envelope information of the current frame. If it is larger than the set threshold,
3. The communication channel decoding method according to claim 2, wherein it is determined that an error is missed. 6. The communication channel decoding method according to claim 2, wherein the error oversight detection step uses a combination of the error oversight detection conditions described in claim 3 to 5 as a condition. 7. A communication channel decoding method for data channel-coded by performing error detection coding on all or some audio information bits and then performing error correction coding, the method comprising: A soft-decision decoding step of soft-decision decoding using soft-decision demodulation data which is a result of quantizing an analog signal obtained by demodulating at a predetermined quantization interval, and performing error detection processing on the soft-decision decoded output. In the error detection step, in the error detection step, when it is determined that there is an error, the number of quantization bits when quantizing the analog signal is set to a larger value, and the soft decision demodulation data is created again. On the other hand, the soft decision decoding step and the error detection step are repeated until it is determined that there is no error in the error detection step, or a predetermined number of repetitions. A communication channel decoding method comprising an iterative decoding step that repeats until a number is reached. 8. A communication channel decoding apparatus for channel-encoded data obtained by performing error detection coding on all or some of audio information bits and then performing error correction coding on the data. And a Viterbi decoder that performs soft-decision decoding using soft-decision demodulation data that is a result of quantizing an analog signal obtained by demodulating the analog signal at a predetermined quantization interval, and performing error detection processing on an output of the Viterbi decoder. An error detector to perform, and in response to an output indicating that there is an error from the error detector, change the quantization interval when quantizing the analog signal to another predetermined set value and repeat the soft decision again. Iterative decoding control for producing demodulated data and repeatedly operating the Viterbi decoder and the error detector until it is determined that there is no error in the error detector, or until a predetermined number of repetitions is reached. A communication channel decoding device comprising: 9. When it is determined that there is no error during the repetition operation, a feature parameter is reproduced from the decoded speech information bit sequence, and whether an error is overlooked using a statistical property of the feature parameter is determined. 9. The communication channel decoding device according to claim 8, further comprising an error oversight detector that determines whether an error has been overlooked and corrects the error detection result to an error when it is determined that an oversight has occurred. 10. The overlooked error detector determines whether the overlooked error has occurred using any one of gain information, pitch period information, and spectrum envelope information or a combination thereof as the feature parameter. 10. The communication channel decoding device according to claim 9, wherein: