JP4319895B2 - Time series signal encoding device - Google Patents

Description

  The present invention relates to a music production field such as music production, storage of acoustic data material, relaying location material, particularly a field of producing high-definition audio with higher quality than a CD, and audio recording using a digital recording medium such as a CD or a DVD. The present invention relates to a data compression coding technique suitable for the field of reproduction, the field of transmission of biological signals in telemedicine, and the like where data modification is hated.

  Conventionally, various methods are used for compression of an acoustic signal. As a method for compressing and encoding an acoustic signal, MP3 (MPEG-1 / Layer3), AAC (MPEG-2 / Layer3), and the like have been put into practical use. Such a compression encoding method makes it possible to handle an acoustic signal as small data, and contributes to the efficiency of data recording and transmission.

  MP3, AAC, and the like as described above are all referred to as lossy encoding methods, and can be efficiently compressed. However, in decoding, the original signal is completely reproduced with a considerable quality degradation. It is not possible.

In order to solve such a problem, a technique for realizing lossless (reversible) compression (see, for example, Patent Document 1) has been proposed.
JP 2000-82199 Gazette

  However, the invention described in Patent Document 1 has a problem that only a compression rate of about 70% can be obtained and a large amount of work memory is required for processing.

  In view of the above points, the present invention can encode a large amount of music data using a small amount of work memory, and can perform signal flat portion compression without imposing a load on the encoding process. It is an object of the present invention to provide a simple time-series signal encoding device.

  In order to solve the above-described problem, an encoding apparatus according to the present invention includes a sample input unit that sequentially reads samples from a sample sequence with respect to a time-series signal including a time-series sample sequence, and a predetermined time-series sample. Encoding mode determination that determines the encoding mode of whether run-length encoding or linear predictive encoding is performed using the immediately preceding buffer, which is a buffer memory for storing the number of samples, and the samples stored in the immediately preceding buffer And run-length coding is selected as the coding mode by the coding mode determination means, the run length is determined by comparing the sample fetched by the sample input means with the latest sample stored in the immediately preceding buffer. Linear predictive coding as a coding mode by a run length calculating means for calculating data and a coding mode determining means When selected, a prediction error calculation means for calculating a linear prediction error using a plurality of samples stored in the immediately preceding buffer, and a sample fetched by the sample input means is written to the immediately preceding buffer, and the sample located at the past Last-minute buffer update means for deleting the previous-length buffer, variable-length coding means for coding the run-length data and the linear prediction error with variable length, and code output means for outputting the coded data And an encoding process control means for controlling each means and outputting the sample of the sample string as a code.

  According to the present invention, when each sample of the time-series signal is read one by one and sequentially encoded, the linear predictive encoding mode is basically used, and the run length mode is encoded depending on the situation. Since it did in this way, it becomes possible to encode a time series signal in real time without being restricted by the limitation of a work memory.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
(1. Time-series signal configuration)
First, a time series signal to be encoded in the present invention will be described. Here, a case of a stereo sound signal having two channels as a time series signal will be described as an example. First, an analog acoustic signal that is a time-series signal is digitized. This may be performed by using a conventional general PCM method, sampling the analog acoustic signal at a predetermined sampling frequency, and converting the amplitude into digital data using a predetermined number of quantization bits. In the present embodiment, the following description will be given on the assumption that a positive / negative code is recorded with a sampling frequency of 44.1 KHz and a quantization bit number of 16 bits. When sampling is performed at a sampling frequency of 44.1 KHz, a sample string composed of 44100 samples per second is formed. FIG. 1 schematically shows a predetermined number of samples from the top. In FIG. 1, the horizontal axis represents time t, and the vertical axis represents amplitude x (t). In FIG. 1, the numbers t−4 to t + 5 indicate the sample numbers of the respective samples, and when sampling is performed at 44.1 KHz as described above, these 10 samples correspond to 1/44410 seconds. Here, the sample number and time are used synonymously. This is because the sample numbers are assigned in ascending order according to the change in time, and a value obtained by multiplying the time by the sampling frequency becomes the sample number. Further, the line segment extending from each sample number in FIG. 1 indicates the amplitude value. This amplitude value is −32768 to 32767 when a positive / negative code is recorded with a quantization bit number of 16 bits as described above. Will take the value.

(2.1. Configuration of encoding apparatus)
The encoding apparatus according to the present invention is realized by mounting a dedicated program for performing encoding processing on a computer. By executing the dedicated program, the computer is caused to perform predetermined processing, and the computer is caused to function as an encoding device.

(2.2. Outline of processing)
Next, the processing operation of the encoding device according to the present invention will be described. FIG. 2 is a flowchart showing an outline of the process of the encoding method according to the present invention. First, one sample is read for each channel from the head (step S10). In the case of a stereo sound signal composed of two channels as in this embodiment, one sample is read alternately for each channel. The read sample is stored in the immediately preceding buffer which is a buffer memory for temporarily storing the immediately preceding several samples. Since up to 5 samples are recorded for each channel in the immediately preceding buffer, the read samples are recorded in order from sample numbers 0 to 4.

  When 5 samples are stored in the previous buffer, other processing is started. First, it is determined what state the encoding mode of this apparatus is (step S20). Specifically, it is determined whether the encoding mode is a linear predictive encoding mode or a run-length mode. In this apparatus, two linear prediction encoding modes and a run length mode are prepared as encoding modes. In the initial state, the linear prediction encoding mode is set. For this reason, first, the process proceeds to step S30. In the case of the linear predictive coding mode, it is determined whether the latest two samples (that is, the fourth and fifth samples from the beginning in the initial state) among the five samples stored in the immediately preceding buffer are the same. . If they are the same, the mode shifts to the run length mode. On the other hand, if they are not the same, the process of the linear predictive coding mode is performed.

(2.2.1. Linear predictive coding mode)
In the linear predictive coding mode, inter-channel correlation processing (step S40), linear prediction error calculation processing (step S50), polarity processing (step S60), and variable length coding processing (step S70) are performed. Variable length bit strings obtained by variable length coding are sequentially output for recording. Thereafter, the current sample read in step S10 is stored in the previous buffer (step S80). Since the sample recorded in the immediately preceding buffer is fixed at 5 samples in this embodiment, the oldest sample is deleted instead of storing the current sample. For example, when samples 0 to 4 are recorded, the current sample 5 is stored and the sample 0 is deleted. As a result, the samples 1 to 5 are stored in the previous buffer. When the current sample is stored in the previous buffer, the process returns to step S10 to read the next sample. Details of inter-channel correlation processing, linear prediction error calculation processing, polarity processing, and variable length coding processing will be described later. While the coding mode is the linear predictive coding mode, the read sample is processed in the loop of Step S10 to Step S80.

(2.2.2 Transition from linear predictive coding mode to run-length mode)
In the state of the linear predictive coding mode (the loop of step S10 to step S80), when the values of the latest two samples match in the immediately preceding buffer in step S30, the coding mode is changed from the linear predictive coding mode, The mode is switched to the run length mode (step S90). As a result, the encoding process is performed from the linear predictive encoding mode to the run-length mode.

(2.2.3. Run length mode)
In the run length mode, first, it is determined whether or not the run is continued (step S100). Specifically, after reading, the value of the current sample not yet stored in the previous buffer is compared with the value of the latest sample among the five samples stored in the previous buffer. If they are the same, it is determined that the run is continuing, and the run length is counted up (step S110). Specifically, the current run length is increased by “1”. Subsequently, the read current sample is stored in the immediately preceding buffer (step S80). Then, returning to step S10, the sample input means reads the next sample. In this state, since the encoding mode is the run-length mode, in step S20, it is determined that the run-length mode is set, and the process proceeds to step S100. While the encoding mode is the run length mode, the read sample is processed in the loop of steps S10, S20, S100, S110, and S80.

(2.2.4. Transition from run-length mode to linear predictive coding mode)
In the run length mode (the loop of steps S10, S20, S100, S110, and S80), in step S100, the current sample that is not stored in the previous buffer and the latest sample among the five samples stored in the previous buffer are displayed. If the values are not the same, it is determined that the run is interrupted, and the run-length information so far is variable-length encoded (step S120). Various variable-length coding methods can be used here, but it is desirable to use a known melcode method. Subsequently, the coding mode is switched from the run-length mode to the linear predictive coding mode (step S130). As a result, the encoding process is performed by shifting from the run-length mode to the linear predictive encoding mode. Subsequently, the difference between the current sample not stored in the immediately preceding buffer and the value of the latest sample among the five samples stored in the immediately preceding buffer is calculated (step S140). Originally, for samples other than the signal flat part where the run is continuous, it is desirable to calculate the linear prediction error using a plurality of past samples, but at this point, since the past plurality of samples is a signal flat part, Effective prediction error cannot be calculated. Therefore, difference processing is performed as a simple method instead of linear prediction error calculation. After this, the linear predictive coding mode loop is entered from step S60.

  If it is a normal acoustic signal, it is relatively rare that the values of consecutive samples are exactly the same, so the processing is most often performed in the linear predictive coding mode. The processing is performed in the run length mode. When the state of the acoustic signal sample changes from the signal flat part to the signal flat part, the mode shifts from the linear predictive coding mode to the run length mode, and the state of the acoustic signal sample changes from the signal flat part to the signal flat part. If it is other than the part, the run length mode is shifted to the linear predictive coding mode. As described above, since the signal flat part is shifted to the run-length mode and encoded, it is possible to particularly efficiently encode the time-series signal having the signal flat part.

(2.2.5. Inter-channel correlation processing)
Next, details of each process from step S40 to step S70 will be described. In step S40, the difference calculation between channels is performed on the samples of each channel. A processing outline of the inter-channel difference calculation is shown in the flowchart of FIG. First, the current sample is read (step S41). Here, the reading means that the module for the inter-channel calculation processing that performs the processing of step S40 reads the sample. If the read sample is for the L channel, the value of the sample is stored in the reference channel memory and output to the linear prediction error calculation process in step S50 (step S42). On the other hand, if the read sample is for the R channel, the processing after step S43 is performed. Note that samples of each channel are alternately read, so there is no need to determine whether the channel is of the L channel or the R channel. For example, when an L channel sample with a sample number t = 1 is read, the next is an R channel sample with a sample number t = 1, and the next is an L channel sample with a sample number t = 2. Therefore, the process of step S42 and the process after step S43 may be switched alternately.

In the case of the R channel sample, the variables Ao and Ad are compared (step S43). Here, Ao is the accumulation of absolute values of the sample values of the R channel, and Ad is the accumulation of absolute values of the differences between the sample values of the R channel and the L channel. The initial values of both variables Ao and Ad are 0. If Ad ≧ Ao in step S43, the R channel sample is output as it is to the linear prediction error calculation process in step S50 (step S44). Further, the accumulated value Ao is updated as shown in the first equation of [Equation 1] below. Specifically, the absolute value of the R channel sample value e _R is added to the accumulated value Ao.

If Ad <Ao in step S43, the difference from the L channel sample stored in the reference channel memory is calculated, and the difference value is output to the linear prediction error calculation process in step S50 (step S45). Further, the accumulated value Ad is updated as shown in the second formula of [Formula 1] below. Specifically, the absolute value of the difference value e _R -e _L between the R channel and L channel samples is added to the accumulated value Ad.

[Formula 1]
Ao ← Ao + | e _R |
Ad ← Ad + | e _R -e _L |

  Subsequently, one counter C indicating that a pair of samples L and R has been processed is added (step S46). As described above, the counter variable C is a local variable used in the inter-channel arithmetic processing module.

  Subsequently, it is determined whether the counter has exceeded a predetermined number of times (step S47). In this embodiment, this predetermined number is set as 100 times. That is, it is determined whether or not the counter C exceeds 100.

  As a result, if the counter exceeds 100, the accumulated values Ao and Ad are halved (step S48). Specifically, as shown in the following [Equation 2], the variables Ao and Ad that are accumulated values are divided by two. At the same time, the counter C is reset to half. That is, Ao and Ad here are not cumulative errors in a pure sense, but are moving averages of cumulative errors. In the present embodiment, the maximum 100 samples immediately before are accumulated, but the previous ones are processed in half. Thereby, the influence of the sample separated in time is made small.

[Formula 2]
Ao ← Ao / 2
Ad ← Ad / 2
C ← C / 2

  By executing the processes in steps S41 to S48 over all samples in all times in the time-series signal, the values of all samples in the R channel are replaced with the difference values from the L channel. However, as is apparent from the above-described processing, there is a sample in which the R channel sample value is recorded as it is depending on the magnitude relationship between the accumulated values Ao and Ad. In the inter-channel calculation process in step S40, all L channel samples are output as they are.

(2.2.6. Linear prediction error calculation processing)
Subsequently, in step S50 shown in FIG. 2, the value of the sample after the inter-channel calculation is converted into a prediction error value. In this embodiment, an adaptive linear predictive coding method is used. In the present embodiment, a maximum of 4 samples of the same channel are required to obtain a prediction error value of a certain sample. An outline of the prediction error calculation process in step S50 is shown in the flowchart of FIG. Since the prediction error calculation process is performed independently for each channel, the case where the process is performed on a sample of one channel will be described with reference to FIG. When processing a plurality of channels, the same process is performed for each channel. First, a linear prediction error corresponding to each prediction calculation formula is calculated using a plurality of prediction calculation formulas prepared in advance (step S51). Specifically, the following [Formula 3] to [Formula 6] are prepared as prediction calculation formulas for calculating the prediction error of the sample number t.

[Formula 3]
e1 (t) = x (t) -x (t-1) -e1 (t-1) / 2

[Formula 4]
e2 (t) = x (t) -2 * x (t-1) + x (t-2) -e2 (t-1) / 2

[Formula 5]
e3 (t) = x (t) −3 × x (t−1) + 3 × x (t−2) −x (t−3) −e3 (t−1) / 2

[Formula 6]
e4 (t) = x (t) -4 * x (t-1) + 6 * x (t-2) -4 * x (t-3) + x (t-4) -e4 (t-1) / 2

  In the above [Equation 3] to [Equation 6], e1 (t) to e4 (t) are prediction errors in the sample at time t according to each prediction calculation equation, and x (t) to x (t-4) are times. It is an amplitude value from t to t-4.

  “2 × x (t−1) −x (t−2)” in the above [Expression 4], “3 × x (t−1) −3 × x (t−2) + x ( t-3) ”,“ 4 × x (t−1) −6 × x (t−2) + 4 × x (t−3) −x (t−4) ”in [Formula 6] Linear prediction component based on ~ 4 samples. Using this linear prediction component and the prediction errors “e1 (t−1) / 2” to “e4 (t−1) / 2” (error feedback component) calculated in the immediately preceding sample, a prediction error at time t e1 (t) to e4 (t) are calculated.

  Subsequently, the linear prediction error that minimizes the cumulative error, which is the cumulative absolute value of the prediction error value for each prediction calculation formula, is selected as the prediction error of the sample (step S52). Here, the concept of cumulative error is used. Specifically, the accumulated values of the past samples of the prediction error calculated by the respective prediction calculation formulas [Formula 3] to [Formula 6] are set as A1 to A4. Then, a prediction error corresponding to the smallest one of the accumulated errors A1 to A4 is selected. For example, it is assumed that A2 is the smallest among A1 to A4. In this case, the prediction error e2 (t) calculated by [Formula 4] is selected as the prediction error e (t) to be encoded. The selected prediction error e (t) is replaced with the original value x (t) of the sample, and the subsequent processing is performed. In order to distinguish from the original sample, the sample in which the prediction error e (t) is recorded is called an error sample.

  Subsequently, the absolute values of the prediction errors e1 (t) to e4 (t) are added to the accumulated errors A1 to A4 (step S53). Specifically, as shown in the following [Equation 7], the variables A1 to A4 that are accumulated error values are updated. At the same time, each time processing of each sample is performed, a process of incrementing the counter C one by one is performed. Note that the counter variable C is a local variable used in the linear prediction error calculation processing module, and is also used in other processes described later, but these are independent variables in each module.

[Formula 7]
A1 ← A1 + | e1 (t) |
A2 ← A2 + | e2 (t) |
A3 ← A3 + | e3 (t) |
A4 ← A4 + | e4 (t) |
C ← C + 1

  Subsequently, it is determined whether the counter has exceeded a predetermined number of times (step S54). In this embodiment, this predetermined number is set as 100 times. That is, it is determined whether or not the counter C exceeds 100.

  As a result, if the counter exceeds 100, the accumulated error is halved (step S55). Specifically, as shown in the following [Equation 8], variables A1 to A4 that are accumulated errors are divided by two. At the same time, the counter is reset to zero. That is, the variables A1 to A4 here are not accumulated values in a pure sense, but are moving averages of accumulated values, similarly to Ao and Ad in [Expression 1] and [Expression 2]. In the present embodiment, the maximum 100 samples immediately before are accumulated, but the previous ones are processed in half. Thereby, the influence of the sample separated in time is made small.

[Formula 8]
A1 ← (A1) / 2
A2 ← (A2) / 2
A3 ← (A3) / 2
A4 ← (A4) / 2
C ← 0

  By executing the processing of step S51 to step S55 over all samples at all times in the time series signal, the values of all samples are replaced with the target error e (t) from the original amplitude value x (t). become.

(2.2.7. Polarity treatment)
Subsequently, in step S60 of FIG. 2, positive / negative polarity processing of each error sample is performed. As a result of the processing of step S40 and step S50, the value of each sample is replaced with the difference between the R channel value and the difference with the L channel, and the amplitude value is replaced with a prediction error. Remains. Normally, when being calculated by a computer such as a computer, each data is processed in units of 32 bits and expressed using a two's complement expression. This is converted into a positive / negative signed absolute value expression, and the absolute value portion is moved one bit higher, and the positive / negative sign bit is moved to LSB (least significant bit). FIG. 5 schematically shows how the bit structure is converted by the polarity processing in step S60. FIG. 5A shows a bit configuration before processing, and FIG. 5B shows a bit configuration after processing. The reason why the positive and negative code bits are moved to the LSB in this way is to make it easier to detect the bit length of each error sample in the variable-length encoding process in the subsequent step S70.

(2.2.8. Variable length coding of samples)
Next, in step S70 of FIG. 2, a process for converting each error sample into a variable length is performed. The variable-length coding in this embodiment employs a known method generally called Golomurice. Specifically, the bit component constituting one sample is divided into an upper bit component and a lower bit component, the lower bit component is left unchanged, and the upper bit component is a numerical value obtained by decimal conversion of only the upper bit. It is an array in which bit “0” is arranged and separator bit “1” is added at the end. For example, consider an 8-bit bit component “00101000”. At this time, if the lower bit component is 4 bits, the lower bit component is “1000”. Since the upper bits are “0010”, “2” pieces of “0” obtained by decimal conversion are arranged and converted to “001” by adding “1” at the end. As a result, the 8-bit bit string “00101000” is converted into a 7-bit bit string “0011000”. In this embodiment, the bit length of the lower-order bit component that makes the bit component unchanged before and after conversion is made variable for each sample.

  Hereinafter, the variable length encoding process in step S70 will be specifically described. FIG. 6 is a flowchart showing an outline of variable length coding. First, an average bit length Bf that is a moving average of the bit lengths of past samples is calculated (step S71). The average bit length Bf is obtained by dividing the cumulative bit length RB, which is the cumulative value of the past bit length, by the counter C based on the past number of samples. That is, Bf = RB / C. Since the accumulated bit length RB is 0 in the initial state, when processing a sample with t = 1, the bit length Bd (t) of the sample with t = 1 is set as an initial value. The initial counter C = 1 is set.

  Subsequently, the bit length Bd (t) of the sample at time t is calculated (step S72). For samples after t = 2, after calculating the average bit length Bf, the bit length Bd (t) of the sample is calculated. This bit length Bd (t) is easily calculated by converting the bit configuration by the polarity processing in step S60. By converting to the bit configuration as shown in FIG. 5B, the bit length starts from the point where the bit “1” appears at the head in the bit configuration of each sample. Next, the bit length Bv of the changing unit is calculated (step S73). This is calculated by subtracting the average bit length Bf from the bit length Bd (t) of the error sample. Subsequently, the data code is output (step S74). Specifically, “0” is output for the numerical value obtained by decimal conversion of the upper Bv bits, then the separator bit “1” is output, and the lower Bf bits are output as the unchanged part. The code output is performed as recording to an external storage device such as a hard disk or a CD-R. Next, the bit length Bd (t) is added to the cumulative bit length RB (step S75). At the same time, every time each error sample is processed, the counter C is incremented by one. Subsequently, it is determined whether or not the counter C has exceeded a predetermined number (step S76). Here, about 100 is set as the predetermined number. Therefore, it is determined whether or not the counter exceeds 100. As a result, if the counter exceeds 100, the cumulative bit length RB is halved (step S77). Specifically, the variable RB that is the cumulative bit length is divided by two. At the same time, the counter C is halved.

  As described above, encoding with variable bit length is performed for each sample. The variable length error sample obtained by encoding is output and recorded on a recording medium or the like as a variable length bit string.

(3.1. Configuration of Decoding Device)
The decoding device according to the present invention is realized by mounting a dedicated program for performing decoding processing on a computer. By executing this dedicated program, the computer is caused to perform predetermined processing, and the computer is caused to function as a decoding device.

(3.2. Overview of decoding process)
Subsequently, an outline of processing of the decoding apparatus according to the present invention will be described. FIG. 7 is a flowchart showing an outline of processing of the decoding apparatus according to the present invention. First, code data is read in units of a predetermined bit string (step S200). At this time, when the time-series signal is composed of a plurality of channels, a predetermined bit string is read for each channel. In the initial state, since the linear predictive coding mode is set, the sample decoding process (step S230), the polarity restoration process (step S240), and the linear prediction error decoding process (steps) are performed on the read variable length bit string. S250) Each process of the inter-channel decoding process (step S260) is performed to restore the original sample, and after the sample is output (step S270), it is stored in the immediately preceding buffer (step S280). Since up to 5 samples are recorded for each channel in the immediately preceding buffer, the read samples are recorded in order from sample numbers 0 to 4.

  When 5 samples are stored in the previous buffer, other processing is started. First, it is determined what state the encoding mode of this apparatus is (step S210). Specifically, it is determined whether the decoding mode is a linear predictive coding mode or a run-length mode. In the decoding apparatus, two linear prediction encoding modes and a run length mode are prepared as decoding modes. As described above, in the initial state, the linear prediction encoding mode is set. For this reason, even if 5 samples are stored in the previous buffer, the process proceeds to step S53 at first. In the case of the linear predictive coding mode, it is determined whether the latest two samples (that is, the fourth and fifth samples from the beginning in the initial state) among the five samples stored in the immediately preceding buffer are the same. . If they are not the same, the process proceeds to step S54 and the process is continued in the linear predictive coding mode. While the decoding mode is the linear predictive coding mode, the read variable-length bit string is processed in the loop of step S200 to step S280.

(3.2.1. Transition from linear predictive coding mode to run-length mode)
In the state of linear predictive coding mode (loop from step S200 to step S280), in step S220, when the values of the latest two samples match in the immediately preceding buffer, the decoding process from the variable length bit string to the run length is performed. This is performed (step S290). Specifically, the decoding process is performed by the Mel code method. Subsequently, the decoding mode is switched from the linear predictive coding mode to the run length mode (step S300). As a result, the decoding process is performed by shifting from the linear predictive coding mode to the run-length mode.

(3.2.2. Run length mode)
In the run length mode, first, it is determined whether or not the run length is greater than 0 (step S310). If the run length is greater than 0, the run length is counted down (step S320). Specifically, the current run length is decreased by “1”. At the same time, one sample having a sample value given to the run length is output. This sample is output as an audio sample, the sample of the original digital sound signal is restored, and the sample is output (step S270). The output sample is analog-converted and then amplified and output as sound from the speaker. The sample is also saved in the immediately preceding buffer (step S280). Then, returning to step S200, the variable length bit string is read. In this state, since the decoding mode is the run-length mode, it is determined in step S210 that it is the run-length mode, and the process proceeds to step S310. While the decoding mode is the run length mode, the run length is processed in the loop of steps S200, S210, S310, S320, S270, and S280.

(3.2.3. Transition from run-length mode to linear predictive coding mode)
In the state of the run length mode (loop of steps S200, S210, S310, S320, S270, S280), if the run length is 0 or less in step S310, the decoding mode is changed from the run length mode to linear predictive coding. The mode is switched (step S330). As a result, the decoding process is performed from the run-length mode to the linear predictive coding mode. Subsequently, the variable length bit string is decoded into samples (step S340). Specifically, a decoding process for Golomb coding is performed. Subsequently, polarity restoration processing is performed (step S350). Further, an addition process with the immediately preceding sample is performed (step S360). The sample after the addition process is output as a decoded sample (step S270). The sample is also saved in the immediately preceding buffer (step S280).

(3.2.4. Decoding of variable length bit string)
Next, details of each process from step S230 to step S260 will be described. In step S230, it is determined which part of the read bit string is a change part and which part is an invariant part, and a fixed bit length sample is restored.

  The outline of the decoding process to the fixed bit length sample in step S230 is shown in the flowchart of FIG. First, a bit string corresponding to the changing unit is extracted from the bit string of the input code data, and is decoded to a fixed length (step S231). The input bit string is followed by 0 from the beginning and 1 appears. This is because variable length coding was performed according to such a rule at the time of the above coding. Therefore, the bit string when “1” appears is decoded according to a rule reverse to that at the time of encoding. For example, when a bit string “0011000...” Comes, it is determined that the first “1” appears up to “001” as a change part, and two “0” s continue. Decode into “0010” (in case of 4 bits).

  Next, an average bit length Bf, which is a moving average of the bit lengths of past samples, is calculated (step S232). This is exactly the same process as the process of step S71 of the variable length encoding. Subsequently, the bit string of the invariant part is extracted (step S233). This is performed by extracting a bit string corresponding to the calculated average bit length Bf. For example, in the bit string “0011000...” As described above, assuming that up to “001” is the changed part and Bf = 4, the subsequent “1000” is extracted as the invariant part.

  Subsequently, the bit string of the changing part decoded in step S231 and the bit string of the invariant part extracted in step S233 are connected to output a fixed-length sample (step S234). The sample restored to the fixed length is output to the polarity restoration process in step S240 shown in FIG. Next, the bit length Bd (t) is added to the cumulative bit length RB (step S235). At the same time, each time the processing of each sample is performed, a process of incrementing the counter C one by one is performed. Subsequently, it is determined whether or not the counter C has exceeded a predetermined number (step S236). Here, about 100 is set as the predetermined number. Therefore, it is determined whether or not the counter exceeds 100. As a result, if the counter exceeds 100, the cumulative bit length RB is halved (step S237). Specifically, the variable RB that is the cumulative bit length is divided by two. At the same time, the counter C is halved. The processing from step S235 to step S237 is exactly the same as the processing from step S75 to step S77 in the variable length coding of FIG.

  If the sample of fixed length is output by step S230, next, in step S240, the positive / negative polarity of each error sample will be restored. This is the reverse of the processing performed in step S60 of the encoding device. Specifically, the sign bit of positive / negative is moved to the MSB (least significant bit), the absolute value part is moved down by 1 bit, and then the sample value expressed by the positive / negative signed absolute value is replaced with 2's complement. Convert to an expression using. When the state of the bit configuration conversion in step S240 is schematically shown, the bit configuration shown in FIG. 5B is converted to the bit configuration shown in FIG. By the processing in step S240, the bit configuration is easy to perform the arithmetic operation by the computer.

  Subsequently, in step S250, the error sample value in which the prediction error value is recorded is restored to an independent value independent of the past sample value. The outline of the independent sample restoration process is shown in the flowchart of FIG. First, it is determined which prediction calculation formula is to be applied among a plurality of prediction calculation formulas prepared in advance (step S251). As the prediction calculation formula, any one of [Formula 3] to [Formula 6] used at the time of encoding is used. Which one of [Equation 3] to [Equation 6] is selected is based on the accumulated error value corresponding to each prediction equation. The accumulated errors A1 to A4 are the same as those used at the time of encoding, and are all 0 in the initial state.

  In the above [Equation 3] to [Equation 6], e1 (t) to e4 (t) are prediction errors in the sample at time t according to each prediction calculation equation, and x (t) to x (t-4) are times. It is an amplitude value from t to t-4.

  As described above, “2 × x (t−1) −x (t−2)” in [Expression 4] and “3 × x (t−1) −3 × x (t) in [Expression 5]. −2) + x (t−3) ”,“ 4 × x (t−1) −6 × x (t−2) + 4 × x (t−3) −x (t−4) ”in [Expression 6] above. "Is a linear prediction component based on the past 2-4 samples. Using this linear prediction component and the prediction errors “e1 (t−1) / 2” to “e4 (t−1) / 2” (error feedback component) calculated in the immediately preceding sample, the sample e at time t The value of (t) is substituted as any one of e1 (t) to e4 (t) into any prediction formula determined in step S251, and x (t) is calculated (step S252).

  Subsequently, the absolute values of the prediction errors e1 (t) to e4 (t) are added to the accumulated errors A1 to A4 (step S253). Specifically, as shown in [Formula 7] used in the encoding apparatus, the variables A1 to A4 that are accumulated error values are updated. At the same time, each time the processing of each sample is performed, a process of incrementing the counter by one is performed.

  Subsequently, it is determined whether the counter has exceeded a predetermined number of times (step S254). In this embodiment, this predetermined number is set as 100 times. That is, it is determined whether or not the counter exceeds 100.

  As a result, if the counter exceeds 100, the accumulated error is halved (step S255). Specifically, as shown in the above [Equation 8] used in the encoding apparatus, the variables A1 to A4 that are accumulated errors are divided by two. At the same time, the counter is reset to zero. That is, A1 to A4 here are not cumulative errors in a pure sense, but are moving averages of cumulative errors. In the present embodiment, the maximum 100 samples immediately before are accumulated, but the previous ones are processed in half. Thereby, the influence of the sample separated in time is made small.

  By executing the processes in steps S251 to S255 on all the error samples that have been read, the values of all the samples are restored from the original prediction error e (t) to the amplitude value x (t).

  Next, in step S260, a process for restoring the value of the R channel sample from the difference value with the L channel sample is performed. An outline of this inter-channel decoding process is shown in the flowchart of FIG. First, the restored sample is read from the prediction error in step S250 (step S261). If the read sample is for the L channel, the value of the sample is stored in the reference channel memory and is output to the next step S270 (step S262). On the other hand, if the read sample is for the R channel, the processing from step S263 is performed. Note that, as in the case of the inter-channel calculation processing of the encoding device, samples for each channel are alternately read, so there is no need to make a determination. For example, when an L channel sample with a sample number t = 1 is read, the next is an R channel sample with a sample number t = 1, and the next is an L channel sample with a sample number t = 2. Therefore, the process of step S262 and the process after step S263 may be switched alternately.

In the case of the R channel sample, the variables Ao and Ad are compared (step S263). Here, Ao is the accumulation of absolute values of the sample values of the R channel, and Ad is the accumulation of absolute values of the differences between the sample values of the R channel and the L channel. The initial values of both variables Ao and Ad are 0. In step S263, if Ad ≧ Ao, the R channel sample is output as it is to the next step S270 (step S264). Further, the processing of the first equation (Ao ← Ao + | e _R |) of the following [Equation 9] is performed. In step S263, if Sd <So, the process of Ad ← Ad + | e _R | is performed, and then the sum with the sample of the L channel stored in the reference channel memory is calculated, and the sum is calculated in the next step S270. (Step S265). The variables Ao and Ad have the same name because they function in the same manner as those used in the inter-channel calculation process of the encoding device. In particular, in the second formula of [Formula 9], unlike the second formula of [Formula 1], the absolute value of the sample value of the R channel is added to the variable Ad, not the difference between the sample values of both channels. . This is because the decoding apparatus processes the sample of the R channel that has been subjected to the difference calculation process by the encoding apparatus. That is, since the difference value between the R channel and the L channel is recorded, the value of the sample of the R channel of the code data is processed by the second equation of [Equation 9]. The same processing as that of Formula 2 is performed.

  Subsequently, one counter C indicating that a pair of samples of L and R has been processed is added (step S266).

[Formula 9]
Ao ← Ao + | e _R |
Ad ← Ad + | e _R |

  Subsequently, it is determined whether the counter has exceeded a predetermined number of times (step S267). In this embodiment, this predetermined number is set as 100 times. That is, it is determined whether or not the counter exceeds 100.

  As a result, if the counter exceeds 100, the accumulated values Ao and Ad are halved (step S268). Specifically, as shown in [Formula 2] used in the encoding apparatus, the variables Ao and Ad that are accumulated absolute values are divided by two. At the same time, the counter C is reset to half. That is, the variables Ao and Ad here are not cumulative values in a pure sense, but are moving averages of cumulative values, as in A1 to A4. In the present embodiment, the maximum 100 samples immediately before are accumulated, but the previous ones are processed in half. Thereby, the influence of the sample separated in time is made small.

  By executing the processing of step S261 to step S268 over all samples at all times in the time-series signal, the values of all samples of the R channel are restored to the values unique to the R channel from the difference values from the L channel. Will be. Since the L channel samples are not changed in the inter-channel calculation process at the time of encoding, the values are output as they are when the channel is restored at the time of decoding.

  As described above, the sample of the original digital sound signal is restored, and the sample is output in step S270. The output sample is analog-converted and then amplified and output as sound from the speaker.

(4. Near lossless coding)
The lossless type encoding and decoding have been described above. However, a near lossless type can be obtained by changing some processes. The near lossless type will be described below. FIG. 11 is a flowchart showing an outline of processing in the case of the near lossless type encoding method according to the present invention. The flowchart in FIG. 11 is mostly the same as the flowchart in FIG. 2, and the difference is that step S150 and step S160 are added. Further, the processes in step S30 and step S100 are slightly different. For this reason, only the processes of steps S30, S100, S150, and S160 will be described here.

  In step S30, in the case of the lossless type shown in FIG. 2, it is assumed that the latest two samples out of the five samples stored in the previous buffer have the same value, but the near lossless type shown in FIG. In this case, if the difference between the two latest samples among the five samples stored in the immediately preceding buffer is within the allowable value, it is determined that they match. For example, if the difference is about ± 1, it is set to match.

  Even in step S100, in the case of the lossless type shown in FIG. 2, when the value of the current sample not yet stored in the previous buffer and the value of the latest sample among the five samples stored in the previous buffer are the same, In the case of the near lossless type shown in FIG. 11, the difference between the value of the current sample not yet stored in the previous buffer and the value of the latest sample among the five samples stored in the previous buffer is acceptable. If it is within the value, it is determined that the run will continue. Again, for example, if the difference is about ± 1, it is set to match.

  In step S150, the error is corrected using the prediction error value e (t) calculated in step S50. Specifically, correction for reducing the error is performed using the following [Equation 10].

[Formula 10]
e ′ (t) = [(e (t) + Vn) / K]
K = (2Vn + 1)

  In [Formula 10], Vn represents an integer value of 1 or more. Actually, about 2% of the maximum value that can be taken by the number of quantization bits is preferable. Therefore, in the present embodiment, it is preferable that the maximum value 32767 that can be taken with 16 quantization bits is about 2%, that is, Vn is about 650. In [Formula 10], [] means the maximum integer that does not exceed the value in parentheses, and thus e ′ (t) indicates an integer value. In this way, a reduced error value e ′ (t) obtained by reducing the prediction error value e (t) is obtained.

  Here, when encoding is performed according to the flow of FIG. 11, the value at which the current sample is decoded differs from the original value in the range of maximum Vn for each sample. If the original value of the current sample is stored in the previous buffer as it is as shown in FIG. 2 in the subsequent step S80, the error from the original value of the value to which the next sample is decoded is expanded to a range of 2Vn. Thereafter, each time the decoding process for each sample is repeated, the error increases. Therefore, the current sample correction process of step S160 is inserted immediately before step S80, and in this step S160, when the value of the current sample is stored in the immediately preceding buffer in the subsequent step S80, the original value of the current sample is stored. Instead, correction is performed so that the current sample becomes a value to be decoded when it is decoded by a method described later. Specifically, in the case of the linear predictive coding mode, correction is performed such that the original value x (t) of the current sample is converted to x ′ (t) in [Formula 11].

[Formula 11]
x ′ (t) = x (t) + e ′ (t) · K−e (t)
K = (2Vn + 1)

  In the case of the run length mode, the current sample is corrected to the same value as immediately before, and the current sample is stored in the previous buffer with the corrected value.

(5. Near lossless decoding)
Next, near-lossless decoding for decoding code data obtained by the near-lossless encoding will be described. FIG. 12 is a flowchart showing an outline of processing in the case of the near lossless type decoding method according to the present invention. The flowchart of FIG. 12 is mostly the same as the flowchart of FIG. 7, and the difference is that step S370 is added. Therefore, only the process of step 370 will be described here.

  In step S370, the prediction error value e (t) is restored using the following [Equation 12] with respect to the restored reduction error value e ′ (t) in step S240.

[Formula 12]
e (t) = e ′ (t) · K
K = (2Vn + 1)

(6. Coding / decoding system using coding device and decoding device)
As described above, the encoding device and the decoding device according to the present invention have been described. However, the encoding device and the decoding device can be connected to each other so as to be able to perform data communication. Specifically, the encoding device is provided with a function of transmitting code data to the decoding device, and the decoding device is provided with a function of receiving code data transmitted from the encoding device. These are realized by installing data communication equipment and software in each computer constituting each encoding device and decoding device.

  By connecting the encoding device and the decoding device in the encoding / decoding system via a general communication line, the code data compression-encoded by the encoding device can be decoded and reproduced almost in real time by the decoding device. As a result, when streaming audio data is distributed, it is possible to transmit a large amount of data without considering the capacity of the communication line.

It is the figure which showed typically the sample sequence of the digital time series signal made into the processing target in this invention. It is a flowchart which shows the process outline | summary of the encoding apparatus of the time series signal which concerns on this invention. It is a flowchart which shows the outline | summary of the correlation process between channels (S40). It is a flowchart which shows the outline | summary of a linear prediction error calculation process (S50). It is a figure which shows the mode of a polarity process. It is a flowchart which shows the outline | summary of a variable-length-coding process (S70). It is a flowchart which shows the outline | summary of the decoding process of the code data obtained by the encoding apparatus of this invention. It is a flowchart which shows the outline | summary of the decoding process (S230) to a fixed length sample. It is a flowchart which shows the outline | summary of the decoding process (S250) from a prediction error. It is a flowchart which shows the outline | summary of the decoding process (S260) between channels. It is a flowchart which shows the process outline | summary of the near lossless encoding by the encoding apparatus of this invention. It is a flowchart which shows the outline | summary of the decoding process of the code data by which the near lossless encoding was carried out.

Claims (10)

An encoding device that compresses the amount of information so that the sample sequence can be reproduced with respect to a time-series signal composed of time-series sample sequences,
Sample input means for sequentially reading samples from the sample sequence;
A previous buffer that is a buffer memory for storing a predetermined number of samples in time series, and
Using a sample stored in the immediately preceding buffer, encoding mode determination means for determining an encoding mode for performing run-length encoding or linear predictive encoding;
When run-length encoding is selected as the encoding mode by the encoding mode determining means, the run length is determined by comparing the sample fetched by the sample input means with the latest sample stored in the immediately preceding buffer. A run length calculating means for calculating data;
When linear prediction encoding is selected as the encoding mode by the encoding mode determination unit, a prediction error calculation unit that calculates a linear prediction error using a plurality of samples stored in the immediately preceding buffer;
A previous buffer update unit for writing a sample taken in by the sample input unit to the previous buffer, and deleting a sample located in the past in the previous buffer;
Variable-length encoding means for encoding the run-length data and the linear prediction error with a variable length;
Code output means for outputting the encoded data;
Have a, a coding control means for code output samples of the sample series by controlling the respective means,
The time-series sample string is composed of a plurality of channels, and the sample input means inputs one sample for each channel, and the immediately preceding buffer, the encoding mode determination means, the run length calculating means, the prediction error calculating means, wherein the immediately preceding buffer updating unit, wherein the variable length coding means coding a time-series signal, wherein der Rukoto performs processing in sample units as many as the number of the channel apparatus. In claim 1 ,
An inter-channel operation means is provided in the preceding stage of the prediction error calculation means, and data obtained by performing a predetermined operation between samples between the channels is passed to the prediction error calculation means. Encoding device. An encoding device that compresses the amount of information so that the sample sequence can be reproduced with respect to a time-series signal composed of time-series sample sequences,
Sample input means for sequentially reading samples from the sample sequence;
A previous buffer that is a buffer memory for storing a predetermined number of samples in time series, and
When the sample stored in the immediately preceding buffer is used to determine the encoding mode for performing run-length encoding or linear predictive encoding, and the encoding mode is set to run-length encoding If nothing is set and linear prediction encoding is set, and the latest two samples among the samples stored in the previous buffer have the same value, the encoding mode is switched to run-length encoding. Encoding mode determination means for initializing the run length data to 0 ;
When run-length encoding is selected as the encoding mode by the encoding mode determining means, the run length is determined by comparing the sample fetched by the sample input means with the latest sample stored in the immediately preceding buffer. A run length calculating means for calculating data;
When linear prediction encoding is selected as the encoding mode by the encoding mode determination unit, a prediction error calculation unit that calculates a linear prediction error using a plurality of samples stored in the immediately preceding buffer;
A previous buffer update unit for writing a sample taken in by the sample input unit to the previous buffer, and deleting a sample located in the past in the previous buffer;
Variable-length encoding means for encoding the run-length data and the linear prediction error with a variable length;
Code output means for outputting the encoded data;
Encoding processing control means for controlling each of the means to code-output samples of the sample sequence;
A time-series signal encoding apparatus comprising: An encoding device that compresses the amount of information so that the sample sequence can be reproduced with respect to a time-series signal composed of time-series sample sequences,
Sample input means for sequentially reading samples from the sample sequence;
A previous buffer that is a buffer memory for storing a predetermined number of samples in time series, and
Using a sample stored in the immediately preceding buffer, encoding mode determination means for determining an encoding mode for performing run-length encoding or linear predictive encoding;
When run-length encoding is selected as the encoding mode by the encoding mode determining means, the run length is determined by comparing the sample fetched by the sample input means with the latest sample stored in the immediately preceding buffer. A run length calculating means for calculating data;
When linear prediction encoding is selected as the encoding mode by the encoding mode determination unit, a prediction error calculation unit that calculates a linear prediction error using a plurality of samples stored in the immediately preceding buffer;
A previous buffer update unit for writing a sample taken in by the sample input unit to the previous buffer, and deleting a sample located in the past in the previous buffer;
A first variable length coding means for coding a variable length for the run length data,
A second variable length coding means for coding a variable length for the previous SL line forms prediction error,
Code output means for outputting the encoded data;
Have a, a coding control means for code output samples of the sample series by controlling the respective means,
The run length calculation means increases the run length data by 1 when the sample taken in by the sample input means is the same as the latest sample stored in the immediately preceding buffer, otherwise The run length data is passed to the first variable length encoding means, and the difference value between the sample fetched by the sample input means and the latest sample stored in the immediately preceding buffer is used as the prediction error value. An apparatus for encoding a time-series signal, characterized in that it is passed to a second variable length encoding means and the encoding mode is switched to linear predictive encoding . An encoding device that compresses the amount of information so that the sample sequence can be reproduced with respect to a time-series signal composed of time-series sample sequences,
Sample input means for sequentially reading samples from the sample sequence;
A previous buffer that is a buffer memory for storing a predetermined number of samples in time series, and
Using a sample stored in the immediately preceding buffer, encoding mode determination means for determining an encoding mode for performing run-length encoding or linear predictive encoding;
When run-length encoding is selected as the encoding mode by the encoding mode determining means, the run length is determined by comparing the sample fetched by the sample input means with the latest sample stored in the immediately preceding buffer. A run length calculating means for calculating data;
When linear prediction encoding is selected as the encoding mode by the encoding mode determination unit, a prediction error calculation unit that calculates a linear prediction error using a plurality of samples stored in the immediately preceding buffer;
The prediction error data is corrected by dividing the prediction error data by a coefficient that is 1 or more to reduce the absolute value of the prediction error data, and the correction is performed on the sample taken in by the sample input means. A prediction error correction means for performing correction for adding a difference from the prediction error data to a value obtained by multiplying the prediction error data by the coefficient, and storing the corrected sample in the immediately preceding buffer;
A previous buffer update unit for writing a sample taken in by the sample input unit to the previous buffer, and deleting a sample located in the past in the previous buffer;
Variable-length encoding means for encoding the run-length data and the linear prediction error with a variable length;
Code output means for outputting the encoded data;
Encoding processing control means for controlling each of the means to code-output samples of the sample sequence;
A time-series signal encoding apparatus comprising: An encoding device that compresses the amount of information so that the sample sequence can be reproduced with respect to a time-series signal composed of time-series sample sequences,
Sample input means for sequentially reading samples from the sample sequence;
A previous buffer that is a buffer memory for storing a predetermined number of samples in time series, and
When the sample stored in the immediately preceding buffer is used to determine the encoding mode for performing run-length encoding or linear predictive encoding, and the encoding mode is set to run-length encoding If no difference is set and linear prediction encoding is set, and the difference between the latest two samples among the samples stored in the immediately preceding buffer is within a predetermined value, the encoding mode is run-length encoding. And encoding mode determination means for initializing the run length data to 0 ,
When run-length encoding is selected as the encoding mode by the encoding mode determining means, the run length is determined by comparing the sample fetched by the sample input means with the latest sample stored in the immediately preceding buffer. A run length calculating means for calculating data;
When linear prediction encoding is selected as the encoding mode by the encoding mode determination unit, a prediction error calculation unit that calculates a linear prediction error using a plurality of samples stored in the immediately preceding buffer;
A previous buffer update unit for writing a sample taken in by the sample input unit to the previous buffer, and deleting a sample located in the past in the previous buffer;
Variable-length encoding means for encoding the run-length data and the linear prediction error with a variable length;
Code output means for outputting the encoded data;
Encoding processing control means for controlling each of the means to code-output samples of the sample sequence;
A time-series signal encoding apparatus comprising: An encoding device that compresses the amount of information so that the sample sequence can be reproduced with respect to a time-series signal composed of time-series sample sequences,
Sample input means for sequentially reading samples from the sample sequence;
A previous buffer that is a buffer memory for storing a predetermined number of samples in time series, and
Using a sample stored in the immediately preceding buffer, encoding mode determination means for determining an encoding mode for performing run-length encoding or linear predictive encoding;
When run-length encoding is selected as the encoding mode by the encoding mode determination means,
When the difference between the sample taken in by the sample input means and the latest sample stored in the immediately preceding buffer is within a predetermined value, the run length data is calculated by increasing the run length data by 1, The sample taken in by the sample input means is corrected so that it is the same as the latest sample stored in the immediately preceding buffer, and the corrected sample is stored in the immediately preceding buffer. Run length data is passed to the variable length coding means, and the variable length coding is performed using a difference value between the sample taken in by the sample input means and the latest sample stored in the immediately preceding buffer as a prediction error value. A run length calculation means for switching the coding mode to linear predictive coding ,
When linear prediction encoding is selected as the encoding mode by the encoding mode determination unit, a prediction error calculation unit that calculates a linear prediction error using a plurality of samples stored in the immediately preceding buffer;
A previous buffer update unit for writing a sample taken in by the sample input unit to the previous buffer, and deleting a sample located in the past in the previous buffer;
Variable-length encoding means for encoding the run-length data and the linear prediction error with a variable length;
Code output means for outputting the encoded data;
Encoding processing control means for controlling each of the means to code-output samples of the sample sequence;
A time-series signal encoding apparatus comprising: In claim 7 ,
The encoding mode determination means does nothing when the encoding mode is set to run length encoding, and when the encoding mode is set to linear predictive encoding, out of the samples stored in the immediately preceding buffer When the difference between the latest two samples is within a predetermined value, the encoding mode is switched to run-length encoding, and the run-length data is initialized to 0. . An encoding device that compresses the amount of information so that the sample sequence can be reproduced with respect to a time-series signal composed of time-series sample sequences,
Sample input means for sequentially reading samples from the sample sequence;
A previous buffer that is a buffer memory for storing a predetermined number of samples in time series, and
Using a sample stored in the immediately preceding buffer, encoding mode determination means for determining an encoding mode for performing run-length encoding or linear predictive encoding;
When run-length encoding is selected as the encoding mode by the encoding mode determining means, the run length is determined by comparing the sample fetched by the sample input means with the latest sample stored in the immediately preceding buffer. A run length calculating means for calculating data;
When linear prediction encoding is selected as the encoding mode by the encoding mode determination unit, a prediction error calculation unit that calculates a linear prediction error using a plurality of samples stored in the immediately preceding buffer;
A previous buffer update unit for writing a sample taken in by the sample input unit to the previous buffer, and deleting a sample located in the past in the previous buffer;
A first variable length coding means for performing variable length sign-reduction with Mel code method for the run-length data,
A second variable length encoding means for performing variable length sign-reduction in the Golomb-Rice method for the previous SL line forms prediction error,
Code output means for outputting the encoded data;
Encoding processing control means for controlling each of the means to code-output samples of the sample sequence;
A time-series signal encoding apparatus comprising: The recording medium which recorded the code data output by the encoding apparatus in any one of Claim 1-9 with respect to the given time series signal.

Images