NO178678B - Circular device for spectral correction and / or post-correction - Google Patents

Description

The invention relates to a circuit device for spectral correction, comprising a spectral analysis circuit which provides a control signal which indicates the possible areas in a frequency spectrum where the predominant components of an input information signal are concentrated, and an adaptive circuit which is controlled by the control signal to impose a variable correction characteristic on the input information signal.

The invention also relates to a circuit device for spectral post-correction and which receives an input information signal and a control signal which indicates the possible areas in a frequency spectrum where the predominant components of the input information signal are concentrated.

In many adaptive A/D and D/A coding systems, the step size used increases with the input noise level. Since quantization noise increases with step size, the quantization noise in such adaptive systems increases with input gss in the signal level, an effect known as noise modulation. The effects of noise modulation are disruptive in many applications, such as high-quality sound reproduction.

It is a characteristic of human hearing that false or unwanted spectral information is much less audible if it is close in frequency to the desired audio signal. If the false energy is far from the desired audio signal frequency, it is much more likely that it is audible. Where the noise level is a function of the input signal level, it is therefore particularly important to reduce noise with frequencies that are far from the desired audio signal.

Conventional noise reduction systems have been used to apply adaptive correction and post-correction to reduce audible noise. A conventional system uses fixed correction for co-copying or raising high-frequency signals and complementary post-correction for counter-coupling or attenuation of such signals. When such correction and post-correction are used to reduce audible noise that increases with the signal level, and ni. if the predominant signals have high frequencies, low-frequency noise will instead be increased.

Fixed correction and post-correction for high frequencies are therefore unsatisfactory for reducing such noise.

A circuit arrangement of a well-known type, called a "slider band", reduces audible high-frequency noise by means of a filter with a variable cut-off or hump frequency. After

as the level of high-frequency signals increases, the filter's corner frequency shifts upward to narrow the band that is raised and cut off. Examples of such circuits can be found in the DS patent documents Re. 28,426, 4,072,914 and 3,934,190.

The shift of the filter's corner frequency depends on both the amplitude and the frequency of the input signal. If such a "sliding band" type circuit is used to reduce audible noise which is a function of the input signal level, low frequency noise may also be increased when the predominant spectral components of the input signal are at very high frequencies. Although such a problem is not as serious as in the case of fixed high-frequency correction and high-frequency post-correction, the "sliding band" type circuit is not entirely satisfactory for reducing noise that increases with signal level.

According to the invention, a circuit device for spectral correction is provided, where the circuit device is of the type indicated at the outset and, according to the invention, is characterized in that the variable characteristic, when the predominant signal components are concentrated in a first frequency range, consists of a sliding high-frequency rise with a corner frequency which shifts upward in response to the control signal to stay above the frequencies of the dominant signal components while leaving the dominant signal components essentially unchanged, while the variable characteristic of the correction, when the dominant signal components are concentrated in a second frequency range with frequencies above the frequencies of the first frequency range , consists of a sliding high-frequency boost with a corner frequency that lies above the frequencies of the predominant signal components, but a damping at the frequencies of the predominant signal components themselves.

According to the invention, a circuit device for spectral post-correction is also provided, where the circuit device is of the type indicated at the outset and according to the invention is characterized by an adaptive circuit which is controlled by the control signal to impose a variable post-correction characteristic on the input information signal, so that when the predominant signal components are concentrated in a first frequency range, the variable characteristic when it comes to post-correction consists of a sliding high-frequency attenuation with a corner frequency that slides upwards in response to the control signal to stay above the frequencies of the dominant signal components, while leaving the dominant signal components in the main case unchanged, while the variable characteristic of the post-correction, when the predominant signal components are concentrated in a second frequency range with frequencies above the frequencies of the first frequency range, consists of a sliding high-frequency attenuation with its corner f frequency over the frequencies of the predominant signal components, but an elevation at the frequencies of the predominant signal components themselves.

The invention is based on the observation that for

to reduce noise by a level that is a function of the input signal level, the various spectral components of the input signal can be changed depending on the spectral composition of the input signal by subjecting the input signal to correction and later post-correction in a complementary manner to recover an output signal that is essentially the same as the input signal. The spectral correction and post-correction circuits of the present invention are particularly advantageous for reducing audible quantization noise in many adaptive A/D and D/A coding systems.

In the circuit arrangement according to the invention, a correction control signal is generated to indicate the spectrum of the input signal and the possible areas in the frequency spectrum where predominant components of the input signal are concentrated. The modifying circuit comprises a device responsive to the correction control signal to apply correction to the spectral components by boosting the components with frequencies above a variable frequency, while the components with frequencies below such variable frequency are held down

(bucking) or is left essentially unchanged. This variable frequency is a characteristic feature of the correction-providing device. It basically slides continuously upwards in frequency when the predominant signal components of the input signal rise in frequency, so that it lies above the frequencies of the predominant signal components.

The recovery circuit has characteristics that are essentially complementary to the characteristics of the change circuit. Such properties of the altering and restorative circuits are particularly advantageous for. reduction of noise., which increases with the signal level as well as with the signal frequency.

The invention will be described in more detail below with reference to the drawings, where fig. 1 shows a graphical illustration of the noise and distortion coming from an ADM encoder-decoder as a function of the step size used by such an encoder-decoder, fig. 2A shows a block diagram of an encoder comprising a pre-processing circuit, illustrating an embodiment of the invention, Fig. 2B shows a block diagram of a decoder comprising a post-processing circuit, illustrating the preferred embodiment of the invention, Fig 3A and 3B show graphical illustrations of respectively pre-correction characteristics of the pre-processing circuit of Fig. 2A and post-correction characteristics of the post-processing circuit of Fig. 2B, Fig. 4 is a block diagram illustrating the preferred embodiment of a portion of the pre-processing circuit of Fig. 2A, Fig. 5 shows a block diagram of a decoder comprising a post-processing circuit, the diagram containing the decoder's system definition to illustrate the preferred embodiment of the invention, and Fig. 6 shows a circuit core for the circuit of Fig. 5 to illustrate the invention.

The circuits according to the invention for changing and restoring spectral composition are particularly suitable for reducing noise which is a function of the signal level.

In the following, a description will be given of an A/D and D/A conversion system whose noise level increases with the signal level. The description will serve as a background which is desirable for understanding the invention. Figs. 2A and 2B show block diagrams of an adaptive A/D encoder and D/A decoder system comprising a pre-processing circuit and a post-processing circuit for changing and restoring the spectral composition of signals, to illustrate the invention. The derivation, transmission and processing of step size information in the coder/cecoder system is the subject of a parallel patent application Mr. 844016" entitled "A/D Encoder and D/A Decoder System". The immediately following discussion regarding step size determination is taken from this parallel application.

The noise and distortion that comes from an ADM encoder-decoder system (codec) depends on the input signal and the step size, both of which are variable. In the following, a codec system that processes a single sine wave is described. As a function of step size, the output noise and distortion will vary as shown qualitatively in fig. 1. In the area marked A, the step size is too large, producing too much quantization noise. In area B, the step size is too small and the system is in overdrive, which produces high noise and distortion. There is an optimal value of the step size for the particular input to tooth marked C. For each short time segment of real sound there is a curve similar to the curve in fig. 1, and an optimal step size. In a conventional, output-controlled ADM system, the step size rarely achieves the optimum value, but stays in range A most of the time, moving into range B during signal transients. The purpose of the present invention is to construct an ADM system which works as much as possible in the area C, and which works so that the delta modulator is fully loaded. This is possible because the step size determination is done in an encoder and is input controlled as described below.

Fig. 2A shows a block diagram of an encoder illustrating an embodiment of the invention. As shown in fig. 2A, an analog audio input signal 12 is passed through a low-pass filter 14 to determine the total audio bandwidth of the analog input signal. Such a bandwidth can typically be 15 kHz. The analog input signal is then passed through a pre-processing circuit lo. The pretreatment circuit's 16" function shall be described below.

After being preprocessed, the analog audio input signal is applied to a step size derivation circuit 18 and to a delay circuit 20. In one particular application, the step size derivation circuit 18 includes a slope detector for detecting the time derivative or slope of the incoming audio input signal. The slope detector then generates a control signal that indicates the step size to be used in the adaptive delta modulator 22. The control signal is limited by a bandwidth limiting circuit 24 and is then supplied to the adaptive data modulator 22. An A/D converter 26 converts the step size control signal into a bit stream of digital signals to convey step-size information. After being time-delayed by the delay circuit 20, the audio input signal is transformed into a bit stream of digital audio signals by the adaptive data modulator 22 in accordance with the step size indicated by the bandwidth limited step size control signal 45. The audio signal bit stream and the step size information bit stream is then transmitted via a medium to a decoder as shown in fig. 2B. In one particular application, the encoder 10 is included in a broadcast station which transmits the audio and step size information bit streams to decoders in user systems. The function of the delay circuit 20 and the bandwidth limiting circuit 24 will be described below after a brief description of the decoder in fig. 2B.

Fig. 2B shows a block diagram of a decoder illustrating the preferred embodiment of the invention. As shown in fig. 2Br, the decoder comprises an adaptive delta demodulator 42 for receiving the digital audio bit stream 30 which is transmitted via the medium, and a D/A converter 44 for

receiving the digital step size information bit stream 28. The D/A converter 44 converts the digital bit stream into an analog two-size control signal which is applied to a bandwidth limiting circuit 46. After being limited in bandwidth, the step size control signal is applied to the adaptive delta demodulator 42. The adaptive delta demodulator 42 generates an analog audio output signal from the audio bit stream in accordance with the bandwidth limited step size control signal from the bandwidth limiting circuit 46. The bandwidth limiting circuit 24 in FIG. 2A limits the bandwidth of the step size control signal supplied to the adaptive data modulator 22 r so that the step size cannot change suddenly from one sample to the next. Similarly, the bandwidth limiting circuit 46 limits the bandwidth of the step size control signal supplied to the adaptive delta demodulator 42. Thus, if the transmission medium introduces a bit error in the step size information bit stream, such an error can, after being converted to analog form of the D/A converter 44, do not introduce a large error in the step size in the adaptive delta demodulator 42. The effect of transmission error is thereby reduced. Inexpensive non-precision components can therefore be used to build the converters 26, 44, and the transmission of the step size information bit stream is very tolerant of bit errors.

The generation of a step size control signal with limited bandwidth will require a finite time. To compensate for this time, the delay circuit introduces 20

a time delay, so that the pre-processed analog input signal will reach the adaptive data modulator 22 at a time when the step size control signal for this input signal is available from the bandwidth limiting circuit 24. This is particularly advantageous when there are sudden changes in the slope of the audio input signal.

Although the step size information bit stream

2 8 and the audio bit stream 30 in fig. 2A and 2B are shown to be

transmitted separately, it will be understood that the two bit streams can be transmitted together in a single channel provided that the two bit streams can be distinguished from each other. Similarly, all three bitstreams 28, 30 and 82 can also be transmitted in the same channel instead of in separate channels if they can be distinguished from each other.

By introducing a time delay via the delay circuit 20 which also compensates for the time delay caused by the bandwidth limiting circuit 46 in the decoder, the audio bit stream signals which are the digital representation of a particular analog signal will reach the data encoder when the bandwidth limited step size control signal is available from the circuit 46". In this way, the need for a delay circuit in the decoder to compensate for the time delay caused by the bandwidth limitation of the step size control signal in the decoder is eliminated, and the decoder circuit is simplified. This is particularly advantageous in lowering the cost of user decoder equipment.

The present invention will now be discussed. The pre-processing and post-processing circuits 10, 96 and other associated circuit components illustrate the invention. As the step size in the A/D and D/A converters is variable, the noise amplitude will be modulated depending on the step size,

and such noise modulation is undesirable in many applications, such as in high quality audio equipment. Noise modulation is reduced by means of the pre-processing circuit 16 and the post-processing circuit 96 when combined with an associated A/D converter 72 and a digital delay 74 in fig. 2A. The pre-processing circuit 16 comprises a spectral analysis circuit 52,

a bandwidth limiting circuit 54, an adaptive pre-correction circuit 56 and a delay circuit 58 all of which are shown in FIG. 2A. The post-processing circuit 96 comprises a D/A converter 76, an adaptive post-correction circuit 78 and a bandwidth limiting circuit 80, all of which are shown in FIG. 2B.

The spectral analysis circuit 52 analyzes the audio input signal to produce a correction control signal.

The produced correction control signal is a function of

only the spectrum of the audio input signal and is essentially independent of the amplitude of the audio input signal. The correction control signal is then limited by the bandwidth limiting circuit 54 and fed to the adaptive pre-correction circuit 56.

The adaptive pre-correction circuit 56 boosts and bucks the various frequency components of the audio input signal by amounts which are functions of the correction control signal. The correction control signal is limited in bandwidth, so that the frequency response of the adaptive pre-correction circuit 56 will not change suddenly from sample to sample. Bandwidth limiting circuit 80 reduces the effect of bit noise introduced by the transmission medium in a manner similar to the previously described bandwidth limiting circuit 46.

Referring to fig. 2A, 2B, the delay circuit 58 introduces a time delay which allows the pre-correction circuit 50 to complete its adaptation before the audio input signal is supplied to the pre-correction circuit 56. The pre-processed audio input signal is supplied to the delay circuit 20 and the adaptive data modulator 22 as described above. The correction control signal from the spectral analysis circuit 52 is transformed into a digital bit stream that carries spectral information, by means of the A/D converter 72, and is delayed by means of the digital delay 74 with a time period which is essentially equal to the delay period of the delay circuit 20.

When comparing the timing relationship between the audio and spectral information bitstreams, an audio signal in the audio bitstream has been delayed by the delay circuits 58, 20, while the corresponding spectral information signal for such an audio signal has been delayed by only the digital delay 74. Thus, the net effect is that the audio is delayed relative to the corresponding spectral information with the time delay caused by the delay 58, so that the spectral information reaches the decoder of Fig. 2B and the post-processing circuit 96 before the corresponding audio signals and at the right time to change the amplitudes of the various frequency components of the audio signal in a manner that is complementary to the change in the adaptive pre-correction circuit 56". The additional conditions for complementarity are discussed below. The time delay introduced in the encoder due to the circuit 58 compensates for the time delay caused by bandwidth limiting of the spectral information in the bandwidth limiting circuit 80 in the decoder.

If the audio and the corresponding spectral information were synchronous, so that they reached the decoder and the post-processing circuit on the same tick, the time delay caused by the bandwidth-limiting circuit 80 would cause the audio signal to reach the adaptive post-correction circuit 78 before the bandwidth-limited post-correction control signal was available to control of the post-correction' A delay circuit would then be required in the decoder to delay the sound so that the sound would reach the post-correction circuit 78 at the correct time. By introducing a timing offset between the sound and the corresponding spectral information in the encoder in the manner described above, the need for a delay circuit in the decode route is eliminated, and the costs for the decoder are reduced.

One of the purposes of the coder-decoder system see fig. 2A and 2B is to transmit an analog audio signal via a medium so that the analog audio output signal recovered after the transmission is substantially the same as the input audio. To achieve this purpose, the adaptive data modulator 22 in the encoder of FIG. 2A and the adaptive delta demodulator 42 are essentially complementary. to each other. Moreover, the step size control signals that are supplied to the modulator 22 and the demodulator 42 are essentially the same and are supplied to the modulator and the demodulator with essentially the same timing relationship with respect to the corresponding audio signals, so that the modulation and demodulation used are complementary. In other words, if the step size control signal for an audio signal is applied at a time t before or after the audio signal reaches the modulator 22, the step size control signal must also reach the demodulator 42 mainly at the time t before or after the audio signal reaches the demodulator. This ensures that the modulation and demodulation used are essentially complementary. Likewise, the pre-processing and post-processing circuits are essentially complementary to each other. The correction and post-correction control signals are essentially the same and have essentially the same timing relationship in relation to the sound, for the application of pre-correction and post-correction so that the pre-correction and post-correction used are essentially complementary.

After the above-mentioned conditions for complementarity are generally fulfilled, however, the encoder-decoder system is very tolerant of timing errors. Then step size-re! sas precorrection and postcorrection control signals can only change slowly in the encoder and decoder, the adaptive modulation and demodulation together with the precorrection and postcorrection applied by the encoder and decoder can only change slowly. A mismatch in the timing relation of the above-mentioned type and of the order of magnitude of a few percent of the bandwidth limitation rise times will thus not cause the modulation and demodulation to deviate significantly from being complementary. Likewise, a mismatch in timing relation

of such a magnitude does not cause the applied pre-correction and post-correction to deviate significantly from being complementary.

The characteristics of the adaptive pre-correction and post-correction circuits 56, 78 are shown in fig. 3A and 3B. It will be understood that the particular frequencies and amplifications in fig. 3A, 3B and in the subsequent description are only for illustration purposes, and that the characteristics or properties of the circuits 56, 78 are not limited by these. In some respects the pre-correction and post-correction characteristics are similar to the well-known type of "sliding band" circuits which reduce high frequency noise by means of a filter with a variable corner frequency. As the signal level increases, the filter corner frequencies of such "sliding band" circuits continuously slide upward to narrow the band that is raised and cut off. As mentioned above, there are examples of such circuits in US patent documents Re 23 426, 4 0 72 914 and 3 934 190.

The pre-correction characteristic of circuit 56 also has a variable frequency which is shown at 86a, 88a, 90a,

92a, 94a and 96a of the respective pre-correction characteristic curves 86, 88, 90, 92, 94 and 96 in fig. 3A. Post-corrections - the curves 84'-96' in fig. 33 are complementary to the respective curves 84 - 96 and also have variable frequencies 86a' - 96a'. These variable frequencies also shift continuously as a function of the input audio signal. Unlike the "sliding band" circuits, however, the continuous shift is determined, not by the level of high-frequency signals, but by the spectral content of the input lye signal in a manner to be described below. In the above "sliding band" circuits, the signal components with frequencies higher than the variable low frequency are boosted (or cut), and the signal components with frequencies lower than the corner frequency remain unchanged. While signals with frequencies higher than the variable frequency are also raised or amplified by the circuit 56 as shown in fig. 3A,

for each of the curves 90 - 96, there is a spectral region where the signals are kept down. Similarly, there is a spectral range for each of the post-correction curves 90' - 96' in which the signals are raised. The detailed characteristics of the circuit 56 will be described below.

For the purpose of description, it is first assumed that the predominant signal components in the input audio signal are concentrated in a certain area of the frequency spectrum. When the audio input signal comprises energy with mostly low and medium frequencies, e.g. concentrated in the frequency range below 500 Hz, the adaptive precorrect ion circuit 56 uses the response curve marked 84, so that only signals with frequencies above 500 Hz are raised. The predominant signals with frequencies below 500 Hz remain essentially unchanged. In other words, there exists a high-frequency rise with a minimum corner frequency of 500 Hz. This minimum corner frequency could alternatively be 100 0 Hz. When the audio signal from the adaptive partial modulator 42 reaches the adaptive post-correction circuit 78, the high-frequency components of the quantization noise will be reduced by the adaptive post-correction circuit 78 which will have a characteristic 84' which is complementary to the curve 84 as shown in fig. 3A and 3B. High-frequency noise above 500 Hz is thereby reduced sufficiently that audible noise modulation is greatly reduced. Noise with lower and medium frequencies below 50 0 Hz is masked by the signal.

As the frequency of the audio input signal rises so that the predominant signal components are concentrated between approx. 500 Hz and 2 kHz, the correction control signal from the spectral analysis circuit 52 causes the frequency response of the adaptive correction circuit 56" to slide from the curve 84 to the curve 86" or 88, i.e. the corner frequency one slides upwards. Such a dynamic effect of the adaptive pre-correction circuit prevents unwanted increases of the step distortion travel in the adaptive data modulator, but still allows the later, complementary post-correction to reduce noise with frequencies above the input signal's frequencies. Low frequency noise is not yet an audible problem.

The frequency responses for curves 84, 86, 88 (ie, sliding high frequency rise) are satisfactory for noise reduction when the predominant spectral components in the input audio signal are below 2 or 3 kHz. Noise with frequencies above these predominant spectral components is reduced as described above. Noise with lower frequencies is masked by the signal. When the predominant spectral components of the input audio signal are at higher frequencies (e.g. above 3 kHz), such sliding lift or gain responses may no longer be satisfactory for noise reduction, as low and mid-frequency noise is no longer masked by the signal. Under these signal conditions, the effect of high frequency boosting would be to increase the step size used in the adaptive delta modulator 22 and the adaptive delta demodulator 42, resulting in an increase in broadband quantization noise. The later, complementary high-frequency cutoff would not reduce the low-frequency part of this increased noise. Low-frequency noise would thus be modulated by changes in the high-frequency components of the input audio signal. Under such conditions, it is desirable to reshape the high-frequency boost of the adaptive pre-correct ion circuit 56 for the spectral range where the predominant signal components in the input audio signal are concentrated to a section, such as the depressions or "dips" shown as parts 90, -. 92b, 94b, 96b of respective curves 90, 92, 94 and 96 in fig. 3A. Therefore, as the frequencies of the predominant spectral components of the input audio signal rise, the frequency response of the adaptive correction circuit 56 would slide past the curves 84, 86

and 88 to curves 90, 92, 94 and 96.

When the predominant signal components are concentrated in high frequencies, such as around 5 kHz, high-frequency noise is masked around 5 kHz- Noise at even higher frequencies may not be masked, and it may be desirable to reduce such noise while reducing low-frequency noise on the front described way. The curves 90, 92, 94 and 96 with frequencies above the variable frequency thus preserve the form of a high-frequency ledge. As shown in fig. 3A, curves 84, 86, 88 tend toward the same fixed gain (eg, 20 dB) at high frequencies. If it is not clearly shown in fig. 3A, the curves 90-9 6 also tend towards the same fixed gain at even higher frequencies. The complementary post-correction curves 84' - 96' which correspond to respective pre-correction curves 84 - 96 are shown in fig. 3B and has variable frequencies 86a' - 96a' which are essentially the same as the frequencies of the pre-correction curves. The post-correction curves 90' - 96' have peaks 90b' - 96b' which correspond to the dips' 90b - 96b in the pre-correction curves in fig. 3A.

The total effect of curves 90 - 96 can now be described. Pre-correction curves with depressions or dips in the spectral ranges of the predominant signal components will reduce the step size and consequently the broadband noise coming from the coder-decoder system. The later post-correct ion peaks 90b', 92b<1>, 94b' and 96b' will select the desired predominant signal components and restore these to their original amplitudes. The post-correct ion will also counteract or hold down signals with frequencies above the variable frequencies, to reduce noise at very high frequencies. Thus, the reduced low-frequency noise level is maintained, high-frequency noise is masked and noise with very high frequencies is reduced.

In the preceding description, it has been assumed that the predominant signal components of the input audio signal are concentrated in a certain area of the frequency spectrum. Such an input signal is actually the most critical case. When the signal's spectral components are more distributed, their masking properties cover more of the noise, and the shapes of the precorrect ion curves are less critical. If the signal's spectral components are distributed in two areas of the frequency spectrum, the pre-correction curve will resemble the curve for the case where the spectral components are concentrated in an area between these two areas.

The bandwidth limiting circuits 24, 46, 54 and 80 limit the step size and spectrum of the control signals to lie within bandwidths of a few tens or low hundreds of Hz. The control signals can consequently have rise times of a few milliseconds. The delay introduced by the delay circuits 20, 58 is therefore chosen so that it is essentially equal to the rise times of the control signals as determined by the bandwidth limitation. Suitable values are in the range 5 - 20 ms. The controlling A/D and D/A converters 26, 44, 72 and 76 may be simple delta or delta-sigma modulators and demodulators operating at a few kilobits per second. second. For television sound applications, a suitable value is half the horizontal frequency, approx.

7.8 kHz.

For ease of instrumentation and better control between encoder and decoder, the signal that enters the bandwidth limiting circuit 54 in the encoder 10 can be derived from the information bit stream 82 instead of the output signal from the spectral analysis circuit 52. Such a configuration is shown in fig. 4,. with the adaptive precorrect ion circuit 56, the limiter 54 and the A/D converter 72 rearranged as shown. A local D/A converter 100 converts the digital stage's drying information from the bit stream 82 into an analog correction control signal. Create an arrangement in fig. 4 is particularly advantageous where the A/D converter 72 uses delta-sigma modulation, so that the local D/A converter 100 is already contained in the A/D converter and no additional, local D/A converter will be necessary. Likewise, the step size information supplied to the adaptive delta modulator circuit 22 can be derived from the step size information bit stream 28. This is again advantageous if the A/D converter 26 uses partial sigma modulation.

Instead of using a bandwidth limiting circuit 46 to limit the bandwidth of the step size control signal supplied to the adaptive delta demodulator, the D/A converter 44 may contain the bandwidth limiting. Likewise, the bandwidth limiting circuit 80 can be eliminated if the D/A converter 76 is bandwidth limited in a similar way.

As it is desirable that the effect of a bit error should be a gain error of similar logarithmic magnitude for both large and small step sizes, it is preferable to construct the A/D converter 26 and the D/A converter 44 so that the digital bit stream 2 8 promotes or transfers the logarithm of the step size. Likewise, the spectral information bit stream preferably conveys or transmits the logarithm of the spectral information. In embodiments where logarithmic and exponential circuits are inconvenient, it may be more convenient to transfer another non-linear function of the step size, such as the square root or cube root. Such functions will not give completely uniform gain errors over the system's dynamic range, but the degree of variation will be much smaller than that which emerges from a linear function.

For the same reasons as discussed above for the transmission of sound information, it is desirable to construct a coder-decoder system that transmits step length information and spectral information with low bit transfer speeds and which can be realized at low costs. In selecting the A/D and D/A conversion system for the converters 26, 44, 72 and 76, it is desirable to select a system that allows a low bit rate for the transfer of step size information. This bit rate is preferably small in comparison with the audio data transmission bit rate.The A/D or D/A conversion performed in the converters 26, 72 and 74, 76 may be a conversion according to many systems, including PGM, delta modulation, or delta-sigma -modulation. Even if a PCM system requires a low bit rate, expensive converters must be used, so it is not desirable to use PCM in the converters. Delta-sigma modulation requires a somewhat higher bit rate (of the order of 5-10 kbit /s) than PCM, but it is realized easily and at low cost.• Furthermore, the bit rate required for delta-sigma modulation is still low compared to the bit rate for transmitting audio data (of the order of 2C0 - 300 kbit/s) Therefore, delta-sigma modulation is used in it ruken execution to be described below. A description of delta-sigma modulation can be found in "Delta Modulation Systems", Pentech Press Limited, London, 1975,

by Raymond Steele.

Fig. 5 shows a block diagram of a decoder system illustrating the preferred embodiment of the invention. The characteristics or properties for most of the circuit blocks are defined for the system in fig. 5. The system is particularly suitable for user use. The adaptive delta demodulator or audio decoder 42 comprises a pulse height modulator 202 and a leaky integrator or leaky integrator 204 . The pulse height modulator 202 multiplies the step size control signal Vss by +1 or -1 in accordance with the audio data bit stream, and supplies the result to the leakage integrator 204. The leakage time constant may be approx. 0.5 ms, which corresponds to a frequency limit of approx. 300 Hz. The integrator integrates the resulting signal to produce an analog audio signal. At frequencies below the frequency corresponding to the leakage time constant, the system is strictly speaking not delta-modulation, but delta-sigma modulation.

As regards fig. 2A, the adaptive data modulator 22 also includes a leakage integrator (not shown) with a cutoff frequency that is approximately the same as the cutoff frequency in the decoder. The step size derivation device 18 may be a slope detector which responds to the pre-processed input audio signal by deriving a control signal indicating the slopes of the signal components of the audio input signal with frequencies above the cut-off frequency, and the amplitudes of the signal components with frequencies below the cut-off frequency.

In the preferred embodiment, the step size or slope data is transmitted by means of delta-sigma modulation and in the form of the logarithm of the desired step drying or slope. The slope data is therefore decoded in the slope decoder 205 by passing through a low-pass filter 206"

(corresponding to the D/A converter 44 and the bandwidth limiting circuit 46 in Fig. 2b) which determines the bandwidth (and consequently the rise time) and the pulsation of the ramp voltage. IN

the preferred embodiment uses a 3-pole low-pass filter which causes the step-size signal V ss to have a rise time of approx. 10 ms corresponding to a bandwidth of approx. 50 Hz. The slope voltage is then supplied to an exponent 208 or anti-logarithmic circuit which can be e.g. a biooiar transistor. If the normalized mean level of the bit stream (or pulse ratio measured over the rise time of the low-pass filter) is written as y,

Vss = VQ e*p ky

where V" and k are constants which are ecmet for the particular realization.

If a practical value of k is 10 ln 2, this definition gives an increase of 6 dB in step size for each increase of 0.1 of y. Since y is limited to a range from 0 to 1, the resulting maximum possible range

of V equal to 60 dB.

pp

The transmission of slope information in logarithmic form reduces the dynamic range that is transmitted in the slope data bit stream from approx. 50 dB to approx. 19 dB, and spreads the effect of bitfeii more uniformly over the dynamic range. Since the Vss <3xunn of the low-pass filter 206 is limited to a bandwidth of about 50 Hz, bit failure results in slow, random amplitude modulation of the output audio signal. The audible disturbance produced by errors in the slope data bit stream, may be negligible Uncorrected bit error rates of up to 1 in 100 or so have been observed to produce almost imperceptible distortion of music or speech.

The low-pass filter thus transforms the digital ones

slope data to analog data and limits its bandwidth. The low pass filter 20 6" therefore performs the functions of both the D/A converter 44 and the bandwidth limiting circuit 46 of Fig. 23. Referring to Figs. 2A, 23 and 5, the delay circuit 20 introduces such a delay that the slope data is received by the filter 20 6 before the corresponding sound data is mated by the pulse-height modulator 202. This time difference compensates for the rise time oå approx. 10 ms of the stage dry electric sess tvresicrn ai et V

ss In this way, the need for a delay circuit in the decoder is eliminated ..•

Fig. 3B illustrates a set of post-correction curves which are complementary to the pre-correction curves of Fig. 3A. There are many ways of synthesizing responses of this nature. The sliding band post-correction circuit 78 shown in FIG. 5, shows one practical realization of the post-correction characteristic of the tic. The system definitions for all circuit blocks in fig. 5 together with a set of values of the constants which give satisfactory results are given below:

s is the complex frequency

TQ = 0.5 ms

is variable so that the variable frequency of the glyceband post-correction f^ is given by

f-, = l/(2TrT7) = f exp kx

Tj = 5 us

T3 = 50 us

T4 = 2 ms

T5 = 25 ys

f = 4 kHz

o

Vq is a scaling factor to match that of the encoder

construct ion.

x and y are the normalized mean levels of their respective bit streams, i.e. the proportion of ones measured over the output time of the 3-point low-pass filter.

k = 10 ln 2 = 6.93

The spectrum decoder 212 comprising the 3-pole filter 214 and the exponent 216 is essentially the same as the healing decoder. It finds the normalized mean level x of the spectrum data input signal that transfers the logarithm to the variable frequency of the desired sliding band post-correction f^ defined above, where f^ is

•different from the variable frequencies 86a - 96a, 86a' -

96a<1> in fig. 3A and 3B. The spectrum decoder generates the exponent or antilogarithm of the mean level and supplies the resulting voltage or current to the slideband postcorrection circuit 78. The correction control signal becomes

less affected by bitfeii in transmission than the slope data control signal.

In delta modulation control systems, the sampling frequency is much greater than the minimum required according to information theory. Non-audio spectral components in the output signal are at frequencies well above the audio frequency band, and only an elementary low-pass filter, such as filter 118, is required.

Fig. 6 is a schematic connection core showing a possible embodiment of the system in fig. 5- As shown in fig. 6, the sliding band post-correction circuit 78 uses a main path 78a with a fixed characteristic in parallel with a further path 78b with a variable characteristic. The variable characteristic of the additional path is controlled by means of the resistance of a variable resistor 252 which in turn is controlled by the correction control signal from the spectrum decoder 212. There is no systematic compression or expansion of the dynamic range. The further path is ultimately determined by the spectrum of the input audio signal.

By limiting the bandwidths of the step size and correction control signals, the characteristics of the partial modulator 42 and the post-correction circuit 78 in fig. 2B only change slowly. Because they have slowly varying or changing characteristics, the delta demodulator and the post-correction circuit are thus made linear or quasi-linear. It makes little difference whether the demodulation is performed before the post-correction circuit or vice versa. This linear or quasi-linear feature of the decoder system is even clearer when it comes to the preferred embodiment of fig. 5. Four processes are performed on the audio bit stream, namely pulse height modulation, leakage integration, sliding band post-correction and fixed post-correction. All four are linear or quasi-linear processes, so they can be performed in any order.

In fig. 5, the high-pulse demodulator 202 can be a relatively simple circuit as it is only necessary to switch the sign of the voltage V depending on the state of the audio bit stream. The modulator 20 2 can thus be made at low cost for user decoders. Multiplication of the audio data by the step size control signal at another point, e.g. however, after the sliding band post-correction, but before the fixed post-correction, can have the advantage of improving the quality of the audio output signal. This may be desirable for such applications as in broadcasting stations and other professional equipment. Although the multiplication will have to be carried out by means of a circuit which is more complicated and therefore more expensive than the type of pulse height modulator which is sufficient for the arrangement of fig. 5, the improvement in quality for professional applications may be well worth the additional cost.. Multiplication of the audio data at a different point'

is permissible because the four processes are functionally linear, as discussed above. All such possible arrangements of the four processes are within the scope of the invention.

Instead of using a single 3-pole filter

in the slope and spectrum decoders 205, 212, it is instead possible to use a two-pole filter if a further, one-pole filter is used to filter the output signal from the exponents 208, 216". The filtering can thus be split into two steps, one before the exponentiation and the second after.Any arrangement of filters can be used as long as the filter for filtering the slope or spectrum data before the exponentiation limits the pulsation in the filtered root once the needle to a few percent of its mean value.

The present invention together with the invention according to the aforementioned parallel application No. 84401S reduces the transmission bit rate of the audio bit stream to a rate comparable to or somewhat less than the bit rate required for a companded PCM system with comparable performance. The bit transfer rates for the coder-decoder system according to the present invention can be in the range of around 200 or 300 kbit per second. second. The transmission of spectral and step size information may require approx. 10 or 20 kbit per second and does not contribute significantly to the total bit transfer rate required for the coder-decoder system according to the invention. The coder-decoder system in fig. 2A and 2B, however, retain the advantages of the delta modulation systems. The invention reduces and in many applications eliminates the disturbing effects of bitfeii. Systemec and its components have high fault tolerance. The receiver equipment (decoder) is cheap. The system is efficient with regard to the use of channel capacity, so that more excess capacity will be available for flexibility to add additional channels, or more bandwidth will be available for other channels, such as video signals. The transmission equipment (encoder) does not require special supervision or the use of non-complementary signal processing.

It will be clear to those skilled in the art that the principles described above are applicable not only to adaptive delta modulation, but also to other adaptive A/D and D/A coding systems, such as delta-sigma- modulation, delta-modulation with double integration, and ?CM systems with variable reference voltages.

Although the invention has been described in connection with the processing and transmission of audio signals, it will be understood that it can just as easily be used for the processing and transmission of other signals. The preceding description of circuit designs and methods is only illustrative of the invention, and various changes with regard to arrangements or other details of method and realization may lie within the scope of the following claims.

Claims (17)

1. Circuitry for spectral correction, comprising a spectral analysis circuit (52) which provides a control signal indicating the possible regions in a frequency spectrum where the predominant components of an input information signal are concentrated, and an adaptive circuit (56) which is controlled by the control signal to impose a variable correction characteristic of the input information signal, CHARACTERIZED IN that the variable characteristic, when the predominant signal components are concentrated in a first frequency range, consists of a sliding high-frequency rise with a corner frequency that slides upward in response to the control signal to stay above the frequencies of the predominant signal components while the predominant signal components are left essentially unchanged, while the variable characteristic of the correction, when the predominant signal components are concentrated in a second frequency range with frequencies above the frequencies of the first frequency range, consists of a sliding high yf frequency boost with a corner frequency that lies above the frequencies of the predominant signal components, but an attenuation at the frequencies of the predominant signal components themselves. 2. Circuit arrangement according to claim 1, CHARACTERIZED IN that when the predominant signal components are detected to be concentrated below a first predetermined frequency in the first frequency range, signal components are raised above the first predetermined frequency, when the predominant signal components are detected to be concentrated below a second predetermined frequency, but above the first predetermined frequency, signal components with frequencies above the frequencies of the predominant signal components are raised, the second predetermined frequency being substantially at the lower end of the second frequency range, and when the predominant signal components are detected to be concentrated above the second predetermined frequency , signal components with frequencies above the frequencies of the predominant signal components are raised, and the predominant signal components are attenuated. 3. Circuit arrangement for spectral post-correction and which receives an input information signal and a control signal indicating the possible regions in a frequency spectrum where the predominant components of the input information signal are concentrated, CHARACTERIZED BY an adaptive circuit (78) which is controlled by the control signal to impose a variable post-correction characteristic on the input information signal, so that when the predominant signal components are concentrated in a first frequency range, the variable characteristic, when it comes to post-correction, consists of a sliding high-frequency attenuation with a corner f frequency that slides upwards in response to the control signal to stay above the frequencies of the predominant signal components, while the predominant signal components are left essentially unchanged, while the variable characteristic of the post-correct ion, when the predominant signal components are concentrated in a second frequency range with frequencies above the frequencies of the first frequency range, consists of a sliding high-frequency attenuation with its corner frequency above the frequencies of the predominant signal components, but a boost at the frequencies of the predominant signal components themselves. 4. Circuit arrangement according to claim 3, CHARACTERIZED IN that when the predominant signal components are detected to be concentrated below a first predetermined frequency in the first frequency range, signal components above the first predetermined frequency are attenuated, when the predominant signal components are detected to be concentrated below a second predetermined frequency, but above the first predetermined frequency, signal components with frequencies above the frequencies of the predominant signal components are attenuated, the second predetermined frequency being substantially at the lower end of the second frequency range, and when the predominant signal components are detected to be concentrated above the second predetermined frequency , signal components with frequencies above the frequencies of the predominant signal components are attenuated, and the predominant signal components are boosted. 5. Circuit arrangement for spectral correction or post-correction according to claim 1, 2, 3 or 4, CHARACTERIZED IN THAT when the predominant signal components are detected to be concentrated below the first predetermined frequency, the signal components below the first predetermined frequency remain substantially unchanged, and when the predominant signal components are detected to be concentrated below the second predetermined frequency but above the first predetermined frequency, the predominant signal components and signal components with frequencies below the frequencies of the predominant signal components, essentially unchanged. 6. Circuit arrangement for spectral correction or post-correction according to claim 1, 2, 3 or 4, CHARACTERIZED IN THAT the spectral components of the input signal with frequencies below the frequencies of the predominant signal components are left essentially unchanged. 7. Circuit arrangement for spectral correction or post-correction according to claim 1, 2, 3 or 4, CHARACTERIZED IN THAT the sliding corner frequency has a predetermined minimum value regardless of how low the predominant frequency range of the input signal is. 8. Circuit arrangement for spectral correction or post-correction according to claim 7, CHARACTERIZED IN THAT the minimum corner frequency is approx. 500 Hz. 9. Circuit arrangement for spectral correction or post-correction according to claim 7, CHARACTERIZED IN THAT the minimum corner frequency is approx. 1000 Hz. 10. Circuit arrangement for spectral correction or post-correction according to one of claims 1-9, CHARACTERIZED IN THAT the second predetermined frequency is approx. 2 kHz. 11. Circuit arrangement for spectral correction or post-correction according to one of claims 1-10, CHARACTERIZED IN THAT when the dominant frequency range of the input signal lies in the second frequency range, the correction characteristic (the post-correct ion characteristic) remains a high-frequency range with elevation (damping) above the predominant frequency range of the input signal, and the components of the input signal in the predominant frequency range is subjected to an attenuation (boost). 12. Circuit arrangement for spectral correction or post-correction according to one of claims 1-11, CHARACTERIZED IN THAT the control signal is bandwidth limited so that it changes slowly in relation to the information signal. 13. Circuit arrangement for spectral correction or post-correction according to claim 12, CHARACTERIZED BY a delay circuit (58) which delays the input signal to the adaptive circuit (56) to compensate for the slow response of the control signal. 14. Circuit arrangement for spectral post-correction according to claim 3, CHARACTERIZED IN THAT the post-correction control signal indicates the normalized mean level x of the spectral information, and that the post-correction used is defined by the relation: where s is the complex frequency <T>x = f0/ (2rc exp kx) T2 is approx. 5 us T3 is approx. 50 us fD is approx. 4 kHz, and k is a constant with a predetermined value. 15. Circuit arrangement according to claim 14, CHARACTERIZED BY the fact that k is equal to approx. 6.93. 16. Use of a spectral correction circuit according to one of claims 1, 2, or 5-13, where the circuit is arranged in front of an analog/digital converter (18, 20, 22, 24) which introduces quantization noise that increases with the signal level. 17. Use of a spectral post-correction circuit according to one of claims 3-15, where the circuit is arranged after a digital/analog converter (42, 44, 46) and reduces quantization noise in the digital input signal to the converter.