Method and System for Translating Digital Signal Sampled at Variable Frequency

The present invention relates generally to data compression techniques, and more particularly to a method and system for translating digital signals sampled at variable frequency.

Many attempts have hitherto been made to reduce the amount of information to be stored in a recording medium or transmitted over telephone lines to a distant end. A known data compression technique includes a method whereby the amplitude of analog signals is logarithmically compressed, and a method known as delta modulation in which the differential component of the analog signal is detected for transmission and the signal is integrated for recovery at the receiving end. In both of these known methods, the analog signal is sampled at a constant frequency which is at least twice the highest frequency of the analog signal to prevent foldover distortion. However, due to the constant sampling frequency quantum noise occurs in the known data compression system over the bandwidth of the recovered signal.

According to a data compression system, as shown and described in Kitamura, U.S. Pat. No. 4,370,643, commonly assigned with the present invention, the sampling interval is determined by the amount of deviation of the original signal from the previously sampled analog value so that the deviation is smaller than a predetermined value of the ratio of the original to the sampled value. While this data compression system is satisfactory under certain circumstances, a disadvantage is that details of the original waveform are not satisfactorily recovered.

The present invention provides a data compression system which eliminates the noise problem involved in data compression. According to this system, an analog signal is sampled at intervals inversely variable as a function of the instantaneous frequency of the original analog signal and a sampling datum representing the sampling interval is paired with the associated digital sample for recording or transmission.

The invention claimed in the appended claims is concerned with a method and system for translating into the original analog signal a digital signal. The digital signal contains a series of data sets, each including (1) sampling datum indicating a sampling interval varies inversely as a function of the instantaneous frequency of said analog signal and, a digital sample indicating the magnitude of said analog signal at the sampling interval.

According to the invention, during decoding or playback mode the digital sample is divided by the sampling datum to derive a quotient which indicates the slope of the signal to be recovered. The quotient is integrated by an integrator to provide interpolation between successive sampling points, so that the original signal is approximated by a plurality of line segments.

More specifically, a microcomputer is provided to store the digital signal and read out each digital sample and its associated sampling datum into first and second digital-to-analog converters, respectively. The outputs of the first and second DA converters are applied to a divider for carrying out the division and thence to an integrator for integrating the output of the divider. A delay time is introduced in the instruction steps of the microcomputer to allow the divider and the integrator to complete the operations--before the next digital sample and sampling datum are read out of the memory. A low-pass filter is additionally connected to the output of the integrator to remove unwanted harmonic components.

The present invention will be described in further detail with reference to the accompanying drawings, in which:

FIG. 1 is a block diagram of a data compression system according to a first embodiment of the invention;

FIG. 2 is a waveform diagram useful for describing the operation of the first embodiment;

FIGS. 3a and 3b together, are a flowchart describing the steps of instructions of the first embodiment executed by the microcomputer of FIG. 1 when the system is operating in a recording mode;

FIGS. 4a and 4b together, are flowcharts of instruction steps which the microcomputer executes when the system is operating in a playback mode;

FIG. 5 is a block diagram of details digital-to-analog converter and interpolator of FIG. 1;

FIG. 6 is a waveform diagram associated with the first embodiment;

FIGS. 7a and 7b together, are a flowchart of a second embodiment of the invention of instruction steps to be executed by the microcomputer operating in a recording mode;

FIG. 8 is a waveform diagram associated with the second embodiment;

FIG. 9 is a flowchart of instructions to be executed according to the second embodiment by the microcomputer when operating in a playback mode;

FIG. 10 is a circuit diagram of details of the digital-to-analog converter of FIG. 1 and a variable frequency low-pass filter constructed according to the second embodiment; and

FIGS. 11a and 11b together are a flowchart of the microcomputer according to a modified form of the second embodiment.

Referring now to FIG. 1, there is shown a data reduction system constructed according to a first embodiment of the present invention. The data reduction system generally comprises a microcomputer 10 and a recording-playback apparatus 20. The apparatus 20 may comprise a tape recording-playback section and a control board 20a having manual controls by which command signals are fed to the microcomputer 10.

A voice signal from a microphone 1 or any other source is applied to a low-pass filter 2 where the frequencies higher than 4000 Hz are suppressed. The output of the low-pass filter 2 is fed to an analog-to-digital converter or PCM encoder 3 which is synchronized with a clock supplied from a time base 4 which forms part of the microcomputer 10. The AD converter 3 samples the signal at intervals ts, FIG. 2, or at a frequency of 8000 Hz and converts the sampled value into a digital sample having an 8-bit code. As will be detailed hereinbelow, the microcomputer 10 is programmed to receive the digitized signal at clock intervals and load it into a buffer memory M1 having a memory capacity of 512 bytes which forms part of the microcomputer and samples the stored digitized signal at longer intervals than ts, determined by the programmed instructions, for transfer to a read-write memory M2 to reduce the quantity of the data bits to be supplied to the recording-playback apparatus or external memory 20. The read-write memory M2 having a capacity of 64 kilobytes stores the sampled digital signals prior to further transfer to the apparatus 20.

During playback modes, digital signal from the apparatus 20 is fed to the read-write memory M2 at clock intervals and thence to a digital-to-analog converter or PCM decoder 5 which is clocked by the time base 4. The output of the DA converter 5 is an analog representation of the data sample and coupled to an interpolator 6. The interpolator 6 provides interpolation between successive analog samples. The interpolated analog signal is further processed by a low-pass filter 7 and delivered to a loudspeaker 8.

The control board includes switches R, P and E. The switch R is operated to initiate recording operation, the switch P is used to initiate playback operation, and the switch E to terminate either of the recording and playback operations. When switch R is operated, a clock signal is supplied to the AD converter 3 and the microcomputer 10 initiates the programmed instructions.

According to a first embodiment of the invention, the microcomputer 10 is programmed to perform the steps as described in various blocks of FIGS. 3a and 3b. These steps are interrupted at periodic clock intervals by an interrupt subroutine 100 to read digital samples from the AD converter 3 into the buffer memory M1. The program begins with a block 101 when the switch R is operated by resetting a data set counter D. In block 102, an 8-bit zero-crosspoint interval counter T and a sampling counter S of the microcomputer 10 are reset to zero. The zero-crossing interval counter T is used to measure the interval between successive zero crossover points of the input analog signal. A digital sample is read from a 512-byte buffer memory M1 in block 103. The counter T is incremented by "1" in response to the reading of each digital sample from the memory M1, block 104. The zero crossing interval T is measured by a program loop formed by block 105 in which the digital sample just read from memory M1 is checked to see if there is a change in sign bit indicating the occurrence of a zero crosspoint and if not, the program returns to block 103 to read the next digital sample from the memory M1, further incrementing the counter T. This process is continued until a zero crosspoint is detected in block 105. The program now enters a block 106 to determine sampling points for transferring digital samples A.sub.1, A.sub.2, . . . A.sub.N-1 from memory M1 to memory M2. In block 106 the count value T, which is representative of the zero crossing interval from time t.sub.0 to t.sub.N (see FIG. 2), is divided into N segments (where N is an integer greater than unity) and multiplied by integers from 1 to (N-1) to generate a series of sampling data T.sub.1 (=T/N), T.sub.2 (=2T/N), . . . T.sub.N-1 (=(N-1)T/N).

A digital sample having an analog value A.sub.r (r=1, 2, . . . N-1) is addressed by a corresponding sampling data T.sub.r and read out of memory M1 as stated in block 107. The sampling counter S is incremented by "1" in block 108 and its count value is checked in block 109 if it corresponds to N-1, and if not, the program returns to block 107 to read the next digital sample A.sub.r+1. This process is repeated until digital samples A.sub.1 to A.sub.N-1 are read out of memory M1. These digital samples and the sampling code T.sub.1 are paired to form data sets (A.sub.1, T.sub.1 ; A.sub.2, T.sub.1 ; . . . A.sub.N-1, T.sub.1) and loaded into the read-write memory M2, block 110. The 15-bit data set counter D is incremented by "1" in block 111 in response to the loading of each data set to count the total number of data sets stored in memory M2. It is seen that there is a predetermined constant number of digital samples for each of the zero crossing intervals T while the latter interval may vary from instant to instant and the number of samples transferred to memory M2 is much smaller than the total number of digital samples stored in memory M1.

The data set counter D is checked in block 112 to see if it has a full count, and if not, the program returns to block 103 to read the next 1-byte from the memory M1. When a full count is reached in the counter D, all the data stored in memory M2 are transferred to the external memory 20 (block 113). The recording mode is terminated by operating a stop switch E, this condition being detected in block 114.

Playback mode is initiated by operation of a switch P on the control board 20a. When this occurs, the program shown in FIG. 4a begins with a block 201 in which the data-set counter D is reset to zero. The apparatus 20 is clocked to transfer the recorded data sets A.sub.r and T.sub.r from the storage medium to the read-write memory M2 at the same rate as they are recorded. This is done by an interrupt routine 200.

The digital samples and corresponding sampling data, now loaded into the memory M2, are addressed sequentially and transferred to the DA converters DAC1 and DAC2, respectively, by having the microcomputer 10 execute blocks 202 to 206. In block 202, the sampling counter S is reset to zero. A digital sample A.sub.r is read from memory M2 and an amplitude difference .DELTA.Ar between Ar+1 and Ar is obtained and transferred to D/A converter DAC1 (block 203). Sampling code T.sub.1 is transferred from memory M28 to D/A converter DAC2. Sampling counter S is incremented by "1" block 205. The counter S is checked in block 206 to see if it is filled to an N-1 count. Blocks 203 to 206 are repeated until S=N-1. When S=N-1, the program exits to a block 207 to increment the data set counter D by "1", and if not, it returns to block 203 to repeat the above process until S=N-1 is reached.

In order to decode the data now retrieved from the memory M2, reference is made to FIG. 5 in which the details of the decoder of the first embodiment are illustrated. The DA converter 5 comprises a first DA converter section DAC1 and a second converter section DAC2 which are respectively arranged to receive differential amplitude and time data (.DELTA.Ar, Tr) from the microcomputer 10 to convert them to corresponding analog values. The interpolator 6 comprises a divider 6a which receives the analog values of the amplitude and time data from DAC1 and DAC2, and an integrator 6b coupled to the output of the divider 6a.

The differential amplitude signal Ar is divided by the time signal T.sub.1 in the divider 6a to provide a quotient Xr which represents the gradient .theta.r to provide interpolation between successive sample points. The integrator 6b provides time integration of this quotient signal to generate a signal Yr (=Xi dt) so that the slope of its waveform is proportional to the gradient .theta.r and varies from one sampling point to another as illustrated in FIG. 6. Therefore, the output of the interpolator 6 is an envelope of successively interconnected line segments which approximate the waveform of the original analog signal.

Through the filtering action of the low-pass filter 7 undesirable high frequency components of the signal from the interpolator are eliminated before it is applied to the loudspeaker 8.

Returning to FIG. 4a, a delay time is introduced in block 207 to permit the converter 5 and interpolator 6 to process a input signals thereof in a manner just described before the next data set is read out of the memory M2.

The microcomputer checks the data set counter D in block 209 to see if it is filled to a full count and if not, advances to a block 210 to check if stop switch E is operated. The program exits from block 210 to block 202 to repeat the above process to reproduce the digital samples contained in the next data set.

It is seen therefore that there is a substantial reduction in data quantity. The external memory system 20 may be dispensed with if the capacity of the internal memory M2 is sufficient to store the information.

Alternatively, the flowchart of FIG. 4a is modified as shown in FIG. 4b in which the blocks 203 and 204 of FIG. 4a are replaced with blocks 220, 221 and 222. In block 220, the microcomputer reads the digital sample A.sub.r, derives the amplitude difference .DELTA.Ar, and proceeds to block 221 to divide .DELTA.Ar by sampling datum T.sub.1 to obtain gradient .theta.r. In block 222, the gradient datum is transferred to the converter DAC1. In this modification, the DA converter DAC2 and the divider 6a of FIG. 5 are dispensed with and the output of DAC1 is directly connected to the input of the integrator 6b.

According to the sampling theorem, quantum noise occurs if the one-half value of the sampling frequency is lower than the frequency range of the information signal. The data reduction system of the first embodiment, however, is not satisfactory in cases where the zero crosspoints are spaced at such long intervals that the one-half value of the sampling frequency becomes lower than the cut-off frequency of the low-pass filter 7.

FIGS. 7a and 7b are illustrations of a flowchart describing the instructions of the microcomputer 10 programmed according to a second embodiment of the invention which eliminates the shortcoming of the first embodiment.

When the manual switch R is operated, the program begins with a block 301 by resetting a frame counter F to zero. In block 302, the microcomputer resets other counters to zero including a zero-crossing counter Z, a 256-byte sampling counter S1 for counting the number of digital samples forming a frame to be described later, and a second sampling counter S2 for counting the number of digital samples to be transferred from the buffer memory M1 to the read-write memory M2. The zero-crossing counter Z is used to register the number of zero crossover points that occur within the frame interval. The frame counter F serves to count the number of frames that have been formed. In block 303 one byte of digital sample is loaded from the AD converter 3 into the buffer memory M1, which is followed by a block 304 to increment the sampling counter S1 by "1". A zero crossover point is detected when there is a change in sign bit of the loaded 8-bit code that signifies the occurrence of a zero crossover point of the input analog signal. This zero crossover point detection is carried out in block 305. If there is no change in the sign bit, the program returns to block 302 to read the next digital sample incrementing the counter S1 by "1". This is repeated until a zero crossover point is detected in block 305. When this occurs, the zero-crossing counter Z is incremented by "1" in block 306 to advance to a block 307 to check if 256 digital samples have been read from the memory M1, and if not, the program returns to block 302 again to repeat the above process until full count (=256) is reached in the sampling counter S1. The frame counter F is incremented by "1" in block 308.

The fact that the sampling counter S1 has a full count is an index that defines a "frame" Tf as shown in FIG. 8. The count value of the zero crossing counter Z up to this moment is an indication of the number of zero crossover points of the input signal that occur in that frame interval.

According to the second embodiment of the invention, a sampling interval Ts is determined for each frame interval for purposes of transferring N digital samples from the buffer memory M1 to the memory M2. This is done by dividing the number of sampling points (=256) by the number of detected zero crossover points Z which is multiplied by a factor K (where, K is an integer), as stated in block 309. A datum N signifying the number of digital samples to be loaded into the memory M2 is determined in block 310 by dividing 256 by Ts.

Using an address determined by Ts, the microcomputer proceeds to read a digital sample A.sub.r (where r=1, 2, . . . N) from memory M1 and the second sampling counter S2 is incremented by "1". These operations are carried out in blocks 311 and 312. In block 313, the contents of counter S2 are checked if it corresponds to N, and if not, the program returns to block 311 to read the next digital sample Ar+1, further incrementing the counter S2 in block 312 until S2=N is detected in block 313, whereby digital samples A.sub.1 to A.sub.N are read out of the memory M1. Digital samples A.sub.1 to A.sub.N and data F, Ts and N are combined in block 314 to form a data set (A.sub.1 -A.sub.N, F, Ts, N) that is stored in the read-write memory M2. When a full count is reached in the frame counter F (see block 315), the data stored in memory M2 is transferred at periodic intervals to the external memory or recording system 20, block 316. If a full count is not yet reached in the counter F, the program returns to block 302 to repeat the above process with respect to the digital samples which form the next frame and enters a block 317 to check if stop switch E is operated to terminate the recording mode.

If it is assumed that K=2, Z=32 and the frame interval is 32 milliseconds, Ts=256/(32.times.2)=4 clock intervals which equals 0.5 milliseconds (=4/8000) and therefore the number of data samples of each frame is reduced from 256 to 64. If use is made of a low-pass filter having a cut-off frequency of 750 Hz, no noise is in the reproduced signal. If the data F, N and Ts are respectively assigned 2, 1 and 64 bytes and the average number of data samples contained in the frame interval is 32, 68 bytes of information is required for each frame and the 64K-byte read-write memory M2 is able to store a 30-second duration of voiced information.

Referring to FIG. 9, playback mode is initiated in response to the operation of switch P, causing the microcomputer 10 to execute the statements in blocks 400 and 401 by resetting the frame counter F and sampling counter S2 to zero. Input data are loaded from apparatus 20 into memory M2 at clock intervals by an interrupt routine 402. The sample number datum N of a given data set is read from the memory M2 in block 403 and the sampling interval datum Ts of the data set is transferred from memory M2 to the DA converter DAC2 (block 404). A digital sample Ar of the data set is transferred from memory M2 to the DA converter DAC1 (block 405). The sampling counter S2 is incremented by "1" in block 406 after each digital sample is transferred to the DA converter DAC1. A delay time is introduced in block 407 so that the time involved in processing the blocks 405 to 408 corresponds to the sampling interval Ts. In block 408, the count value of the sampling counter S2 is checked if it corresponds to N, and if not, the program returns to block 405 to transfer the next digital sample Ar+1 of the given data set to converter DAC1. Therefore, digital samples A.sub.1 to A.sub.N of the given data set are transferred to the converter DAC1 during each frame interval. The frame counter F is incremented by "1" in block 409. The program goes through blocks 410 and 411 and returns to block 401 to reset the sampling counter to zero to repeat the above process until the frame counter F is filled to a full count or switch E is operated.

FIG. 10 is an illustration of the details of the DA converter DAC2 and a variable frequency low-pass filter according to the second embodiment of the invention. The DA converter DAC2 comprises an analog switch 30, an operational amplifier 31 and a plurality of feedback resistors R1, R2, R3 and R4. The analog switch 30 is responsive to the sampling interval data Ts to selectively couple one or more resistors between the inverting input and output terminals of the amplifier 31, so that the latter has a variable gain according to the sampling interval data. The variable frequency low-pass filter 40 comprises a pair of photocouplers PC1 and PC2 having variable resistance elements VR1, VR2 connected in series between the output of the DA converter DAC1 and the noninverting input of an operational amplifier 41 through high-frequency determining resistors R5 and R8; the inverting input of amplifier 41 is coupled by a capacitor C1 to a junction between low-frequency determining resistors R6 and R7. A capacitor C2 is coupled between the noninverting input of amplifier 41 to ground. Capacitors C1 and C2 are also used to determine the cut-off frequencies of the low-pass filter 40. The output of amplifier 41 is coupled to amplifier 42, having an output fed to the loudspeaker 8. The photodiode elements PD1 and PD2 of the photocouplers are connected from the output of amplifier 31 to ground.

The resistors R2 and R3 are variable resistors which are adjusted so that the cut-off frequency of the low-pass filter 40 may correspond to the sampling interval. As a function of the digital value Ts, the amplifier 30 provides a variable impedance to the photodiodes PD1 and PD2. The impedance values of variable resistors VR1 and VR2 vary their impedance values in response to the varying brightness of the photodiodes PD1, PD2 so that the following relationships are established between sampling clock interval Ts, sampling frequency fs, and the cut-off frequency fc of the filter 40:

It is seen that the cut-off frequency of the filter 40 becomes automatically equal to the one-half value of the varying sampling frequency, and therefore, no quantum noise occurs in the signal applied to the loudspeaker 8.

FIGS. 11a and 11b are a flowchart describing an alternative form of the recording mode of the second embodiment. In interrupt subroutine 500, digital samples are loaded from the AD converter 3 to the memory M1 at clock intervals ts. The program starts with a block 501 in which the frame counter F is rest to zero. Subsequently, the sampling counters S1 and S2 are reset to zero in block 502. A digital sample is read from the memory M1 in block 503 and the first sampling counter S1 is incremented by "1" in block 504. Blocks 505 to 509 describe steps for detecting the number of harmonic components of the digital samples that occur within a frame interval. A technique known as "Fast Fourier Transform" is used for this purpose. While this technique may be used to detect a frequency spectrum of such digital samples and generate therefrom a power spectrum by simulataneouly treating them in a single subroutine, it is preferable that the digital samples of each frame be divided into eight groups of 32 samples each and the FFT technique be applied in respect of each sample group. For this reason, a decision step is provided in block 505 to check if the counter S1 has reached a count of 32, and if so, the program is advanced to block 506 to use the FFT to derive a frequency spectrum from the group of 32 digital samples just read out of the memory M1. A power spectrum is subsequently derived in block 507 from the frequency spectrum. The blocks 502 to 507 are repeatedly executed until power spectrums are derived respectively from eight groups of 32 digital samples when S1=256 is detected in a decision step in block 508.

In block 509 the microcomputer derives a peak or average value of the power spectrum from the power spectrums of the individual groups and advances to block 510 to remove unwanted higher frequency components having spectral values lower than (1/64)th, for example, of the detected peak or average value. In block 511, the highest order of the harmonic components of the power spectrum is detected by checking the power spectrum of which the unwanted components have been removed. This harmonic order value is used in block 512 to determine a corresponding sampling interval Ts from the following relationships:

In block 513, the number of digital samples to be retrieved from the memory M1 is determined by dividing 256 by Ts and the sampling number counter N is set to 263/Ts. Digital samples A.sub.l to A.sub.N are sequentially retrieved from the memory M1 by executing a program loop including blocks 514 to 516: in block 514, a digital sample A.sub.r is read out of memory M1 and in block 515, the second sampling counter S2 is incremented by "1" and in block 516, S2=N? is checked. The frame counter F is then incremented in block 517. A data set F, A.sub.l to A.sub.N, Ts and N is formed in block 518 and loaded into the memory M2. The program returns through blocks 519 and 520 to block 502 to reinitialize the sampling counters to repeat the above process.