Digital audio companding and error conditioning4809274Abstract A system for companding digital audio signal samples processes the digital audio signal samples to correct errors induced by the compression and expansion processes. These errors are calculated prior to compressing the samples. Such errors are calculated in accordance with a predetermined compression process and a predetermined expansion process; and the digital audio signal samples are corrected in accordance with such calculations prior to compression. The compression process includes providing a 3-bit gain word for a block of 70 samples. The gain word is computed in accordance with the position of the most significant "1" bit in the sample(s) having the peak magnitude. Each individual sample is processed in accordance with the block gain word to compress the sample by reducing the number of magnitude bits. The error calculation process includes calculating an error signal by subtracting a calculated reproduced digital audio signal sample from the digital audio signal sample to be compressed, filtering the error signal by noise-spectral filtering and adding the filtered error signal to the next-providing digital audio signal sample to provide a corrected digital audio signal sample for compression. The error calculation process also includes calculating an error value based upon the effect of using a single gain word for compressing a block of samples. The error value is subtracted from the corrected digital audio signal sample prior to compression. The gain word is forward error corrected by (5,1) encoding to enable detection and correction of both single-bit and double-bit errors in transfer by majority voting. The sign bit and the most significant magnitude bits of the compressed samples are forward error corrected to enable detection and correction of single-bit errors in transfer and to enable detection and concealment of double-bit errors in transfer. Claims We claim: Description BACKGROUND OF THE INVENTION
TABLE 1
______________________________________
GAIN COMPRESSED REPRODUCED
MSB WORD MAGNITUDE BITS MAGNITUDE BITS
______________________________________
B 111 BCDEFGH BCDEFGH1000000
C 110 CDEFGHI 0CDEFGHI100000
D 101 DEFGHIJ 00DEFGHIJ10000
E 100 EFGHIJK 000EFGHIJK1000
F 011 FGHIJKL 0000FGHIJKL100
G 010 GHIJKLM 00000GHIJKLM10
H 001 HIJKLMN 000000HIJKLMN1
I-O 000 IJKLMNO 0000000IJKLMNO
______________________________________
Table 1 shows the relationship between the most significant bit position having a "1" bit in the detected peak magnitude sample(s), the computed gain word and the compressed magnitude bits provided in accordance with the compression process performed by the compression unit 20. The magnitude bits in the remaining positions of the binary digital audio signal sample on lines 50 are truncated. Table 1 further shows the corresponding binary values of the reproduced digital audio signal samples provided in accordance with the expansion process performed by the expansion unit 33. Note that in the expansion process a "1" bit representing one-half the value of the least significant magnitude bit of the compressed magnitude bits is appended to the compressed magnitude bits to represent the average value of the magnitude bits that were truncated by the compression process. The output calculation unit 18 provides calculated compressed magnitude bits for each sample on line 53 to the ROM 21. The ROM 21 permanently stores all of the different combinations of the fourteen calculated reproduced magnitude bits corresponding to each possible combination of calculated compressed magnitude bits, and responds to the calculated compressed magnitude bits on lines 53 by providing the fourteen calculated reproduced magnitude bits for the immediately processed sample on lines 54 to the second subtraction unit 22. The subtraction unit subtracts the calculated compressed magnitude bits on lines 54 from the magnitude bits of the digital audio signal sample on lines lines 50 to provide a system output error signal on lines 55. The multiplication unit 23 multiplies the error signal on lines 55 by the sign bit on line 46 to provide an error signal on lines 56 that is filtered by the FIR filter 24. The FIR filter 24 filters the error signal on lines 56 to provide the filtered error signal on line 49 that is added by the second adder 25, as described above. The adder 25 adds the filtered error signal on lines 49 to the next-provided digital audio signal sample on line 47. Thus errors from prior samples are accumulated and a smaller output error is possible, when the output bandwidth of the system is less than the sampling rate F]hd 2/2. The FIR filter 24 processes the error signal on lines 56 by noise-spectral filtering to reduce audibly perceived truncation errors and/or RMS truncation errors when the filtered error signal is added to the next-provided digital audio signal sample on lines 47. The filtering characteristics are determined by the selection of the coefficients of the FIR filter 24. The noise shaping feature allows the system designer to change the spectral content of quantization noise generated by the compression unit 20. Traditionally, preemphasis and deemphasis are used to contour an audio system's noise spectral density to improve the perceived quality. There have always been complaints about the loss of head room due to preemphasis (clipping will occur at lower levels for high frequencies than lower frequencies.) The use of noise shaping to contour the system noise does not produce any such difference in clipping level versus frequency. Preemphasis and deemphasis are nevertheless retained in the preferred embodiment because the subjective effects of bit errors are significantly reduced by the deemphasis. Accordingly, the preemphasis unit 10 contours the spectral density of the input analog audio signal on line 39; and the deemphasis unit 36 (FIG. 2) deemphasizes he reproduced analog audio signal on line 48 to contour the quantization noise spectral density of the reproduced analog audio signal. Such contouring reduces the audibly perceived effect of any bit errors in the reproduced digital audio signal samples. Noise shaping is a method typically used to reduce the number of input or output states required in a D/A or A/D process operated at several times the required Nyquist sampling rate. In the system of the present invention the noise shaping process is applied to slightly oversampled systems. (10 to 20%). There are significant gains in signal to quantization noise ratios for large signals possible with this feature. Fore example, when the RMS error in a 20 kHz bandwidth is measured with the sampling frequency equal to 44 kHz, the output bandwidth equal to 18.7 kHz, and preemphasis and deemphasis applied (50/15 .mu.sec time constants), the gain is 3.1 dB. There are larger subjective gains available by selecting a different criteria for the coefficients of FIR filter 24. By compromising the RMS improvement obtained in a 20 kHz bandwidth by 0.1 dB, the perceived signal to quantization ratio can be improved to 6.0 dB. This is equivalent to one bit of additional accuracy or conversely allows an additional bit of compression for equivalent perceived quality. Another benefit realized with this feature is primarily a subjective advantage. Correlated error components can occur on low slope (low frequency) signals when there is insufficient dither. The correlation of adjacent samples results in inharmonic tones of varying frequency. This is particularly sever for low frequency signals (20-100 Hz) since the inharmonic tones occur around 1 kHz where the human ear is most sensitive. This is more audible and disturbing than equivalent amounts of white noise added to the signal, which is the effect when there is no correlation between adjacent samples. The feedback structure of the FIR filter 24 breaks up correlated signal components by effectively dithering the input audio samples with shaped quantizing noise. The output calculation unit 18 also provides on line 57 an error value that is related to the effect of using a gain word for a block of samples when companding individual samples of the block. As noted above, in the expansion process, a "1" bit representing one-half the value of the least significant magnitude bit of the compressed magnitude bits is appended to the compressed magnitude bits to represent the average value of the magnitude bits that were truncated by the compression process. Depending upon the appropriateness of the block gain word for the individual sample, the effect of appending this particular "1" bit can be quite significant. For example, if the block gain word is "111" and the most significant "1" bit of the sample is in bit position "J", the effect of appending this particular "1" bit upon expansion is the same as adding the binary value of "64" to the individual sample. Thus, for this example, the output calculation unit provides an error value on lines 57 having a binary value of "64". Table 2 shows the calculated error values for different gain words in relation to the bit position of the most significant "1" bit in the individual digital audio signal sample.
TABLE 2
______________________________________
ERROR VALUES
MSB POSITION
GAIN WORD B C D E F G H J
______________________________________
111 0 1 2 4 8 16 32 64
110 0 0 1 2 4 8 16 32
101 0 0 0 1 2 4 8 16
100 0 0 0 0 1 2 4 8
011 0 0 0 0 0 1 2 4
010 0 0 0 0 0 0 1 2
001 0 0 0 0 0 0 0 1
000 0 0 0 0 0 0 0 0
______________________________________
The subtraction unit 19 subtracts the error value on lines 57 from the magnitude bits on line 50 to provide a corrected digital audio signal sample on line 58 for compression by the compression unit 20. The compression unit 20 provides the seven compressed magnitude bits on lines 60. The digital signal is forward-error-correction coded for transfer. The sign bit on line 46 and the three most significant magnitude bits on lines 60 are provided to the (13,8) block encoder 27 together with corresponding sign and magnitude bits on lines 61 and 62 from a paired audio channel. The (13,8) block encoder 27 provides five parity bits for the eight sign and magnitude bits provided thereto and provides the five parity bits together with these eight sign and magnitude bits on thirteen lines 63 to the interleaver and parallel-to-serial conversion unit 28. The (13,8) encoder encodes these eight sign and magnitude bits to enable detection and correction of single-bit errors in the transfer of these eight bits and to enable detection of double-bit errors therein. A moderately exhaustive search was performed to select the code implemented by the (13,8) encoder; and a large group of codes with roughly equivalent distance profiles exits. The cyclic code derivative was selected for ease of implementation, in that it allows an area efficient decoder implementation. The selected code generator matrix is shown in Table 3. The notation for presenting such a matrix is described in "Information Theory and Reliable Communication" by R.G. Gallager (1968).
TABLE 3
______________________________________
: 1 0 1 0 1
:
: 1 1 1 1 1
:
: 1 1 0 1 0
:
: 0 1 1 0 1
I :
: 1 0 0 1 1
:
: 1 1 1 0 0
:
: 0 1 1 1 0
:
: 0 0 1 1 1
______________________________________
I = 8 .times. 8 IDENTITY MATRIX
The error control is as effective, but more efficient than in the prior art. The system described in the above-referenced U.S. Pat. No. 4,608,456 provides single-bit error correction and double-bit error detection and concealment, but requires 4 bits per sample to achieve this. The number of holds generated varies as approximately two times the channel probability bit error rate (P.sub.E). The European Broadcasting Union has described a system which achieves double-bit error detection and single-bit error correction at 5 bits/sample with a hold rate of 78 P.sub.E .sup.2 which is superior to the described in the system of the '456 patent at bit rates less than 2.6.times.10.sup.-2 where both systems operate. ("Specification of the System of the MAC/Packet Family", Tech. 3258-E, European Broadcasting Union, Oct. 1986, Technical Center-Brussels). The error control described herein has 2.5 bits per audio sample, and the same 78 P.sub.E .sup.2 hod rate characteristics. The prior art has roughly equivalent error control in terms of capability, but the efficiency of this system is 1.5 to 2.5 bits per sample more efficient. Overall the compression algorithm is 2 bits/sample superior to prior art. The noise shaping compression process of the present invention is interoperable with .mu.-law or A-law DACs and saves 1 bit per sample. The error control is equivalent to that of the prior art with 1.5 to 2.5 bits saving per sample. The overall result is a 4.5 to 5 bits per sample savings with equivalent quality. Additionally, granularity on low slope signals is reduced. The gain word on lines 45 is provided to the (5.1) encoder 26 together with a gain word on lines 64 for the paired audio channel from which the sign bit on line 61 and the seven magnitude bits on lines 62 are derived. The (5,1) encoder provides each of the six gain word bits five times on line 65 to the interleaver and parallel-to-serial conversion unit 28. The gain words are thereby forward-error-correction coded to enable detection and correction of both single and double errors in the transfer of any bit of a gain word by majority vote processing the five repeated bits. The four least significant bits of the compressed magnitude bits provided on each of lines 60 and 62 respectively are provided directly to the interleaver and parallel-to-serial conversion unit 28 without any forward-error-correction coding. The interleaver and parllel-to-serial conversion unit 28 interleaves the sign bits, the magnitude bits and the parity bits in accordance with the delay pattern set forth in Table 4 so as to provide a Hamming distance of ten between coded bits of the same sample. Thus noise bursts up to the ten samples in duration can be handled. The interleaving of the uncoded least significant magnitude bits reduces RMS noise energy.
TABLE 4
__________________________________________________________________________
SERIAL
ORDERED
POSITION DEINTERLEAVER
INTERLEAVER
ON CHANNEL
TYPE OF BIT DELAY DELAY
__________________________________________________________________________
1 SIGN BIT, LEFT 0 5
2 SIGN BIT, RIGHT 1 4
3 (MS) MAGNITUDE BIT 6, LEFT
2 3
4 (MS) MAGNITUDE BIT 6, RIGHT
3 2
5 MAGNITUDE BIT 5, LEFT
4 1
6 MAGNITUDE BIT 5, RIGHT
5 0
7 (LS) MAGNITUDE BIT 0, LEFT
1 4
8 (MS) MAGNITUDE BIT 0, RIGHT
2 3
9 MAGNITUDE BIT 1, LEFT
3 2
10 MAGNITUDE BIT 1, RIGHT
4 1
11 MAGNITUDE BIT 4, LEFT
0 5
12 MAGNITUDE BIT 4, RIGHT
1 4
13 PARITY BIT 4 2 3
14 PARITY BIT 3 3 2
15 PARITY BIT 2 4 1
16 PARITY BIT 1 5 0
17 MAGNITUDE BIT 2, LEFT
1 4
18 MAGNITUDE BIT 2, RIGHT
2 3
19 MAGNITUDE BIT 3, LEFT
3 2
20 MAGNITUDE BIT 3, RIGHT
4 1
21 PARITY BIT 0 0 5
__________________________________________________________________________
Referring to Table 4, the terms "left" and "right" are used to designate two different audio channels; and the terms "MS" and "LS" refer to most significant and least significant, respectively. One gain word bit is transferred for every forty-nine interleaved sample bits; hence it is not necessary to also interleave the gain word bits. In a preferred embodiment, in which the compressed and coded bits are transferred during the horizontal blanking interval HBI) of a television signal, the interleaver and parallel-to-serial conversion unit 28 provides the bits in the order shown in FIG. 3 in each sequence of three video lines. Referring to FIG. 3, "S-1" indicates set number one in a sequence; the number in parentheses indicates the number o bits from that set, "GW" indicates one gain word bit; and "CB" indicates the video color burst that is typically broadcast during the HBI. Seven complete sets are transferred over the duration of three video lines. Thus a block of seventy coded, compressed digital audio signal samples for a pair of audio channels are transferred during the duration of thirty video lines. During this 30-line duration, thirty gain word bits are transferred thereby providing five repetitions of each of the three gain word bits for each of the two audio channels. In the decoder, as shown in FIG. 2, the deinterleaver and serial-to-parallel conversion unit 30 deinterleaves the transferred coded and compressed sample bits and provides the deinterleaved bits in parallel on lines 67, with the eight coded bits being provided to the (13,8) block FEC decoder 32. The deinterleaver and serial-to-parllel conversion unit 30 also provides the repeated gain word bits on line 68 to the (5,1) majority decoder 31. The deinterleaver delays are set forth in Table 4 above. The (13,8) block decoder 32 detects and corrects any single-bit errors in the set of eight coded bits and detects and conceals any double-bit errors in the set of eight coded bit. Concealment is accomplished by repeating the last correct or corrected paired samples in lieu of the samples in which the detected double-bit errors occur. The (13,8) block decoder 32 provides the eight decoder sign and magnitude bits on lines 69 to the expansion unit 33. The (5,1) majority decoder 31 detects and corrects any single-bit or double-bit errors by majority voting of the five repeated bits for each bit of the gain word, and provides the three gain word bits for each of the two audio channels in parallel on line 70 to the expansion unit 33. The least significant of the compressed magnitude bits were not coded for transfer to the decoder (FIG. 2), whereby they are provided directly to the expansion unit 33 on lines 71. The expansion unit 33 separates the gain words and sign and magnitude bits for the separate audio channels and processes the gain word, sign bit and magnitude bits of an individual sample for a single channel to provide reproduced digital audio signal samples for each of the audio channels on separate 15-bit line sets 51 and 52 respectively. The composition of the reproduced digital audio signal samples provided by the expansion unit 33 is set forth in Table 1, above. When a .mu.-law companding process is utilized, the binary value of "64" is subtracted from the binary signal value of the reproduced digital audio signal sample by the expansion unit prior to providing the reproduced digital audio signal samples on lines 51 and 52 for conversion to analog audio signals by the DACs 34. In each audio output channel, the DAC 34 converts the reproduced digital audio signal samples on lines 51 into an analog audio signal on line 74. Alternatively a companding DAC may be used. A companding DAC combines the expansion and digital-to-analog conversion functions. Companding DAC's for accomplishing either .mu.-lay or A-law expansion are known to those familiar with the digital signal companding art. Such companding DACs are readily available and their use results in savings in manufacturing costs.
|
Same subclass Same class Consider this |
||||||||||