DE3440613C1 - Method for digital transmission of a broadcast radio programme signal - Google Patents

Abstract

For digital transmission of a broadcast radio programme signal, the frequency band of the broadcast radio programme signal is split into sub-bands. The individual sub-bands are differently quantised and are subsequently transmitted using time-division multiplex. In this case, the quantisation of each sub-band is carried out on the basis of the quiescent audibility threshold and/or on the basis of the monitoring threshold which is applicable to the respective bandwidth of the sub-band, and/or on the basis of the amplitude statistics of the broadcast radio programme signal which are applicable to the respective frequency of the sub-band.

Description

Adaptation to the hearing thresholds This requirement leads to the known PCM systems lead to unnecessarily high information flows because the »broadband PCM channel« basically not completely to the "frequency group-wide coded channel" of the Can be adjusted to the ear: Broadband PCM systems generate broadband quantization noise, whose Spectrum becomes highly frequency-dependent sensitivity of hearing is unfavorable. This is illustrated with reference to Fig. 1, where the Frequency curve of the quiet hearing threshold and a hearing threshold of hearing are entered are.

The hatched area in FIG. 1 below the quiet hearing threshold corresponds to the just imperceptible sound pressure level. The spectrum of the empty channel noise a PCM path may therefore have the frequency curve of the quiet hearing threshold to to be inaudible. The energy of the empty channel noise is allowed in the range of high and low frequencies correspondingly higher than in the frequency range 2 ... kHz.

In contrast, broadband PCM links produce approximately white Idle channel noise. The noise power of the broadband PCM link must therefore in the whole Frequency range can be kept so small that the quiet threshold even in the sensitive Frequency range 2 ... 5 kHz is not exceeded.

A PCM link whose idle channel noise is white thus transmits an irrelevant flow of information. It results from the hatched area below the quiet hearing threshold and is around 64 kbit / s. The flow of information in a PCM link can therefore be reduced by 64 kbit / s without loss of quality if that Spectrum of the empty channel noise by suitable coding the course of the quiet hearing threshold is adjusted.

Furthermore, broadband digital companding is not optimal regarding the masking properties of the human hearing. Examples of digital Companding principles are: - Use of non-linear quantization scales, instantaneous value companding; - Use of several linear quantization scales with different level heights and transfer of the scale factor. Block companding (one scale factor per sample block, z. B. 32 samples).

Compared to linear quantization, known methods achieve a Reduction of the information flow by about 100 ... 150 kbit / s, depending on the degree of companding. In return, however, a higher noise power is accepted for high signal levels ("Companding noise"), d. i.e., the quantization noise of the empty channel increases at higher signal levels. The masking effects associated with higher signal levels however, hearing is insufficient for critical program types. All so far Suggested digital companding methods (e.g. 14 on 11 or 14 on 10) produce a companding noise that can be clearly perceptible (»noise plume«), even when using extensive pre-deemphasis (e.g. according to CCITT Rec. J 17).

Broadband digital companding can be achieved without any loss of quality only a small bit rate reduction because of the broadband companding noise the hearing threshold of hearing can very easily exceed.

Fig. 1 shows the eavesdropping threshold, for example when transmitting a Narrow band noise (fm = 800 Hz), and reproduction with 50 dB sound pressure: the sensitivity of the ear is hardly lower in the frequency range above 2 kHz than without a masker. This even applies to low-pass noise, fgr = 800 Hz. One in the range of 2 ... 5 kHz is inadequate Concealment of the companding noise by the useful signal limits the possible Degree of companding strongly. This results from the monitoring threshold curves of the Hear other ways to reduce the flow of information: by doing the Spectrum of the companding noise by suitable coding the course of the listening threshold curves adapts, the degree of companding can be increased considerably without any loss of quality. The attainable level can be achieved in the same way as with the idle hearing threshold adaptation Saving in channel capacity from the area below the respective effective monitoring threshold curve derive.

Hearing threshold adjustment through frequency-dependent quantization The hearing thresholds can be understood as the - frequency-dependent - level of internal auditory noise.

As in a PCM channel, quiet noise (empty channel noise) can be distinguished from signal-dependent noise. The optimal adaptation of a Sound transmission channel to the hearing would therefore be the replica of the internal auditory Coding what cannot be done. Rather, it must be »artificial«, in terms of communications technology Realizable codings are used, which the noise behavior of the hearing in Sine can use a minimal channel capacity. This can be coded in the frequency or be a time domain, fixed or adaptive coding with or without "memory": they can only be optimal if the noise spectrum corresponds to the hearing thresholds.

With regard to the corresponding use of the eavesdropping threshold is to Note that the hearing thresholds within limited frequency bands (frequency groups) forms. The decomposition of the frequency spectrum into frequency bands apparently leads to the courses of the monitoring threshold curves. If one assumes that there is a signal-dependent "Coding noise" occurs, the course of the monitoring threshold curve in principle by the frequency group-wide band limitation of the coding noise explain. The course, which is flatter compared to the frequency group filters, can be seen here the edges, especially the upper edges at high levels, by "crosstalk" explain in the neighboring frequency groups.

The concept of an optimal coding can be derived from this: The coding noise of a channel can be optimally masked by the useful signal, if the coding is not broadband, but frequency group-wide. In particular the coding can be as precisely as required within each (any narrow) bandpass be adapted to the noise behavior of the hearing.

The aforementioned principles of the invention are now to be based on PCM routes to be viewed as.

Although the quantization noise in a PCM system is very flat Has frequency spectrum, can by bandpass-specific quantization or companding the noise adjustment take place. As will be shown, this so-called turns out to be "Bandpass PCM" as effective for taking into account both the quiet hearing threshold as well as the eavesdropping threshold. If one assumes that the quiet hearing threshold is through the The effect of an internal "quiet noise" can result from correspondingly different Dimensioning of the quantizations within the bandpass-limited PCM channels Empty channel noise of the PCM path optimally adapted to the quiet noise of the hearing It is also assumed that the eavesdropping threshold is increased by the action of a internal "coding noise" results, then by correspondingly different Dimensioning of the companding within the bandpass-limited PCM channels das Companding noise of the PCM path optimally matches the coding noise of the hearing be adjusted.

Since the steepness of the resting hearing threshold curve is much smaller are than the steepnesses of the eavesdropping threshold curves, is the required width of the bandpasses determined by the masking of the companding noise. The concealment of the empty channel noise is less critical. Frequency group widths are obviously optimal Band passes that obscure the companding noise to a high degree, and that is why enable high degrees of companding. This is a question of the technical effort.

Adaptation to the amplitude statistics around the modulation range to use a PCM route completely, even at high frequencies, use pre / de-emphasis, which, according to the amplitude statistics, are particularly effective in the range of high frequencies are. The common pre-deemphases reduce the perceptibility of the quantization noise, by reducing the noise level in the high frequency range.

Long-term spectrum In Fig. 2 the relationships are in two examples The quiet hearing threshold and the spectrum of white noise are entered (W.R), which is just subliminal, as well as the resulting spectra according to Evaluation by the "50>" or J17-Pre / Deemphasis ". Decrease the pre-de-emphasis the noise level in the range of high frequencies and thereby shift the noise level, which is just covered by the difference dL, where AL is the quiet threshold distance is. For the 50 Fs pre / deemphasis, ALso = -4 dB, for the J17 pre / deemphasis vlLJ7 = -12 dB, resulting in savings of 21 kbit / s (50 taS) or 64 kbit / s (J17) result.

It can be seen from Fig. 2 that the amount of possible savings of channel capacity in broadband PCM systems solely through the amount of pre-distortion is determined in the 3 kHz range. Only the sensitive area of hearing in this one Frequency range is decisive for whether the quantization noise is excessive or is subliminal.

In principle, however, it is precisely in the area of high frequencies the greatest possible pre-distortion.

The example of 50! Ls-Pre / Deemphasis clearly shows that the Predistortion in broadband PCM is not optimally used to save channel capacity will. Compared to the frequency-independent clip limit of the PCM route is arithmetic a bit rate reduction of 41 kbit / s is possible if the clipp limit changes that has 50 is deemphasis. This reduction is technical through frequency division can be achieved, for example, by bandpass filters having a width of eight thirds in the upper frequency range.

Apparently the frequency-dependent one possible in bandpass PCM systems Quantization can not only be used advantageously for the purpose of noise adjustment, but also for adapting the channel to the amplitude statistics. Bandpass PCM systems can use the amplitude statistics more effectively than broadband PCM systems.

For answering the question of which cliff limit frequency curve is tolerable, the following has to be observed: Like more recent amplitude statistical investigations can show in modern music, especially in electronic music and impulsive Sounds, the 50 ps limit curve and even the 25 Fs limit curve for short-term overloads lead if a modulation meter according to ARD specification 3/6 is used.

In broadband PCM systems, clipping occurs for a short time strong, non-linear distortion with sufficient headroom, i.e. at the expense of reducing the signal-to-noise ratio. For this Basically, a lowering of the clipping limit appears in accordance with the J 17 curve in Broadband PCM systems hardly make sense. However, the disadvantages of (extensive) Pre / deemphasis in bandpass PCM systems lower: As a result of the bandpass limitation the distortion products that occur briefly due to clipping are partially suppressed and much better covered. The clipping noises - depending on the Dimensioning of the band passes - more or less moderated. Especially in With narrow band pass filters, there are significantly fewer levels of control problems that in the range of high frequencies the clipping limit can be effectively reduced allowed. Already octave-wide bandpasses suppress practically all harmonic distortion products, which occur due to clipping.

Short-term spectrum In addition, there is the possibility of the principle To use "variable pre-emphasis", an adaptive emphasis can alleviate the level problems Reduce further if pre- and de-emphasis by inversely transferring the control variable work towards each other. The flow of information required for this is low.

In contrast to broadband PCM systems, bandpass PCM systems this frequency-dependent, short-term adaptation can be carried out very easily. The moment effective frequency response of the pre-emphasis does not have to be through short-term spectral analysis determined and adjusted by appropriate variable filters, but it a simple complementary gain control is carried out in each channel of the bandpass PCM path.

The advantage here is that the one optimized for the hearing threshold adjustment Bandpass division also a favorable bandpass division with regard to amplitude statistics adaptation represents.

Such an adaptive complementary channel gain that depends works from the level of the useful signal can be designed, for example, in such a way that that only in the case of very high useful signal levels in the coder a relatively high channel gain is regulated back (limiter characteristic). In this case, the noise reduction results from the level increase in the coder and the complementary level decrease in the decoder; it is reduced in the case of overdrive according to the limiter characteristic.

The adaptive complementary channel gain can also be used in the Be designed so that in the case of low and medium useful signal levels in the coder the channel gain is increased (dynamic compressor characteristic). In this Case arises from the complementary channel gain in the decoder for low and middle level a compander gain according to the compressor characteristic.

When using adaptive complementary gains in the channels The bandpass PCM path shows another advantage over the bandpass PCM Broadband PCM, because the known characteristics (limiter, compressor, floating point) Can be adapted favorably to the short-term spectrum. The static and dynamic Characteristic values of the channel reinforcements can thus be optimized in a simple way, that compared to broadband systems a significantly higher noise reduction is possible is.

Basic structure of the bandpass PCM in FIG. 3 is a principled one Structure of the bandpass PCM shown. The frequency conversion and bandpass filtering take place analogue for better comprehensibility. In terms of circuitry, there are various digital ones so-called »transmultiplexers« implemented, which convert frequency division multiplex signals (FDM) into time division multiplexed signals (TDM) and vice versa. It is easy to see that a broadband Analog signal of width B in principle as a frequency multiplex signal of width B = > can be understood as Bj. For example, some procedures are described in the The Bell System Technical Journal, October 1976, pages 1069-1085 and May-June 1977, pages 747 to 770 and in the dissertation »Dimensioning of digital TDM-FDM transmultiplexers using the polyphase method «, ETH Zurich Discussed and described in 1979.

On the basis of the basic structure in Fig. 3, it will be explained below: as by choosing the frequency band division (number of center frequencies and width of the Bandpasses) as well as the reduction of the transmission bit rate through the design of the quantization can be achieved without deteriorating the signal quality.

After the bandpass division, all higher-lying channels are frequency-converted, so that the sampling frequency 1> 2 Bi is sufficient in the subsequent PCM conversion. The net information flow 17DM of the time-division multiplex signal is in principle not greater as the net information flow IEDM of the frequency division multiplexed signal, it reduces with a reduction in the signal-to-noise ratio in the individual channels according to the requirements the hearing threshold curves of the hearing are possible.

On the decoder side, the TDM signal is inversely transposed back to the encoder side. As a result of the bandpasses at the channel outputs, the quantization noise becomes each Channel bandpass limited. If the quantization is the same in all channels, the result is The usual (broadband) noise of a broadband PCM line appears at the output of the decoder. If, for example, the resolution is reduced in one of the channels, it increases accordingly the noise power of the bandpass PCM path for the spectral components, which are in the pass band of this channel.

Adaptation of the idle channel noise to the quiet hearing threshold, FIG. 4 shows the quiet hearing threshold in the third octave diagram. The hatched area represents the exploitable Spectrum. A channel capacity of approximately is required for the transmission of this spectrum 60 kbit / s required (limit frequency 15 kHz). The bulk of that capacity results from the high frequency part of the spectrum. That becomes clear in the Representation in Fig. 5. The possible savings in channel capacity are shown Third, calculated according to the relationship CiGewinn = 1 Biaai 3 Bi: bandwidth of the band pass i with center frequency fj aq /: reduction of the signal-to-noise ratio (dB) in the bandpass filter i Der The low-frequency component does not make a significant contribution (because of the low bandwidth Bi of these thirds). With the last three thirds 8, 10, 12.5 kHz including 1/4 of the 16 kHz third, a saving of 48 kbit / s can be achieved, that is 80% of the possible total savings. This amount also arises if the last Octave of the entire spectrum, namely 7.5 ... 15 kHz is taken as a basis: it is with B1, = 7.5 kHz and aqi ', = 19 dB (see quiet hearing threshold) Cc ii, = 48 kbit / s.

As a technically sensible way of saving through adaptation of the empty channel noise at the quiet hearing threshold is offered to halve the Frequency band. This solution uses 80% of the possible savings and requires for the sampling only a common sampling frequency of f = 16 kHz. On this way 48 kbit / s can thus be saved without any loss of quality.

A common sampling frequency for all channels of the bandpass PCM offers technical advantages especially with word and block synchronization.

Adjusting the Companding Noise to the Eavesdropping Threshold Below it is explained why the signal-to-noise ratio is determined solely by choosing the width Bi of the bandpass filter i a is determined, which is necessary to achieve complete concealment.

The occlusion curves are when they are over instead of over frequency the Bark scale can be applied, regardless of the pitch z. Above about 500 Hz is z # lgf, i.e. H. even at high frequencies we can see roughly the same curves assume as with medium frequencies, see Fig. 1. The concealment effect of a Maskierers behaves above 500 Hz in the 1st approximation independent of the frequency position of the masker.

It should also be noted that the very steep lower flanks the listening thresholds are also independent of the sound level of the masker (above 500 Hz), the slope is constant at around 80 ... 90 dB / oct.

To conceal a bandpass limited companding noise To be able to estimate, the worst-case example according to Figure 6 is assumed. The illustration shows the threshold curve for fM = 1 kHz, sound pressure level L, T, = 60 dB. If the upper limit frequency of the bandpass filter fo = 1 kHz, then lies the lower limit frequency of the companding noise at fu = fo - Bi. In the sketch if B, third octave is assumed, here the level of the companding noise must be about 25 dB below the masking level ("critical signal-to-noise ratio" for one third wide Rush).

Mathematically, the critical signal-to-noise ratio aqi results from the steepness of the lower edge of the monitoring threshold curve and from the maximum height of the monitoring threshold curve. If one takes the "hearing thresholds of a tone, hidden by narrowband noise" as a basis, the steepness of the lower edge for frn = 1 kHz is 85 dB / oct. and the maximum L - L Tlaxst 5 dB (see Fig. 6). This results in the critical signal-to-noise ratio fov fui: upper and lower cut-off frequency of the bandpass filter i As a first approximation, the relationship is independent of the frequency and the sound pressure. This no longer applies if the monitoring threshold curves for masking sine tones are used as a basis: The steepness of the lower edge is approx. 30 dB / oct. for LM = 30dB and approx. 110dB / Oct. for LM = 90 dB; in addition, L .u - L Tniax is 11 dB. However, from the relationship given above, the values for the critical signal-to-noise ratio aqi are only slightly too smaller for sine levels LM> 60 dB.

This representation is based on the theoretically worst case. In practice the ratios are more favorable for two reasons.

1. The transmission of a single line exactly at the upper limit frequency foj of the band pass practically does not occur. The line would be roughly in the middle of the pass band, the critical signal-to-noise ratio would already be halved will.

2. The completely noise-free transmission of a single line and Reproduction with sound pressure levels> 60 dB is hardly possible, even for electronic music relevant.

For the worst-case estimate, the critical signal-to-noise ratio should therefore be assumed with With qj quantization levels, this value has a signal-to-noise ratio opposite to. Equations (3) and (4) can be used to directly determine the bit rate required to mask the companding noise. 1dl = 33.2 lg (foi) +0.52 [bit] fui sample For the estimation it is assumed: idi = 32ig (;) + 1 [simple 1 (6) This results in the following table of values, for example: Table 1: Bandwidth Bj: 1 third 2 thirds 3 thirds Critical signal-to-noise ratio aqj: 25 dB 45 dB 65 dB For masking required Idqi 4.1 bit 7.4 bit 10.6 bit sample sample sample sarnple sample You can see the favorable effect of a narrow bandpass, but also the unfavorable conditions with broadband companding. If the bandwidths are greater than 4 thirds, the eavesdropping threshold no longer covers the companding noise, but merely reduces the interference effect.

At this point I would like to point to a favorable situation pointed out. In practice it can be assumed that the spectral components in Range of high frequencies, for example from around 7.5 kHz, only as overtone or Noise component of an overall sound occur. Even electronic musical instruments hardly produce any sounds with only high-frequency components. Therefore, the area becomes higher Frequencies the concealment of the companding noise in the bandpass filter i mainly through Useful signal components take place in the bandpass filter 1 - 1. Because the upper edges of the eavesdropping threshold run much flatter, the bit rate required for masking is in the range high frequencies practically smaller than estimated according to equation (6).

With a view to a favorable adaptation of the bandpass PCM channel to the ear canal should be recorded: - The bit rate required for masking is practically only dependent on the bandwidth, but practically not on the Bandpass center frequency and the monitoring volume.

- A bandpass PCM system that maintains the critical signal to noise ratio, does not produce any audible companding noise even at high monitoring levels.

- A bandpass PCM system that falls below the critical signal-to-noise ratio, can generate audible companding noise if the companding noise is above the quiet hearing threshold.

If the critical noise level is below the quiet hearing threshold, so hides the quiet hearing threshold. In this case either the degree of companding unnecessarily high or the bandpass filter unnecessarily narrow. For dimensioning the bandpass PCM So in principle the maximum listening volume is important.

Adaptation of the bandpass PCM channel to the ear canal For determination a favorable adaptation of the bandpass PCM channel to the ear canal becomes the »signal-to-noise ratio of the ear canal «determined depending on the frequency range and in the known S / N diagrams shown. To do this, the monitoring volume must be set. It is believed, that with full modulation of the PCM path the sound pressure level is 90 dB. The nominal level S0 = 0 dB of the PCM path results in a nominal sound pressure level of L 50 = 90 dB. It is further assumed that the headroom is 10 dB.

With a view to the frequency curve of the resting hearing threshold, the adjustment determined for three »typical« center frequencies: f ,, l = 0.8 kHz, f "2 = 3.2 kHz, f", 3 = 10 kHz. Fig. 7 shows the assumed relationships in the listening area. Registered are also the eavesdropping thresholds for full modulation and the critical signal-to-noise ratios for one, two and three thirds (1T, 2T, 3T), see Table 1.

It can be seen that the critical signal-to-noise ratio at a lower sound pressure level, So with lower modulation of the PCM path, lie below the quiet hearing threshold can. For example, the required signal to noise ratio of a 10 kHz tone is not more equal to the critical signal-to-noise ratio (aqio kf (,) 3T = 65 dB if the level of the 10 kHz tones becomes smaller than 10 dB under full modulation: this is already effective here "Hiding" the quiet hearing threshold. If the level is reduced further, the signal-to-noise ratio may be in the 3-third octave-wide band-pass, it falls proportionally with the signal level.

This course of the signal-to-noise ratio of the hearing is the course of the required signal-to-noise ratio of the PCM channel. As usual, it can be in an S / N diagram and are derived from the relationship between the nominal sound pressure level and the nominal level the requirements for the band-pass-limited channel with regard to idle channel noise directly and companding noise.

The 450 straight lines resulting from The nominal sound pressure level and the quiet hearing threshold result in that they are frequency-dependent. Furthermore, the critical signal-to-noise ratios for the bandwidths are 1T, 2T and 3T (1, 2 and 3 thirds) entered, which result from the listening threshold; they are frequency independent. The curves describe exactly the "signal-to-noise ratio of hearing" (SS r (i.e. the distance of a tone from the hearing threshold in the range of a band pass): In the range of lower sound pressure levels L s, the distance to the (quiet) hearing threshold increases of the band-pass limited noise proportionally with Ls. In the area of high sound pressure levels L s (depending on the bandpass width) remains the spacing of the bandpass-limited noise constant from the hearing threshold.

As is known, such an (S / N) curve also characterizes PCM links with non-linear, especially logarithmic quantization (e.g. 13-segment compander characteristic): In the area of low signal levels S the distance (S / N) PCM increases proportionally with S on; in the range of high signal levels (S / N) P (M remains constant to a first approximation.

The optimal quantization for the bandpass filter i of the bandpass PCM path can therefore be taken directly from the (SAV), r curves. Fig. 8 makes first important correlations clearly: 1. The position of the horizontal straight line indicates the maximum required signal-to-noise ratio (SSV) pc, ç.

It depends solely on the bandwidth of the bandpass channel: According to equation (3): 2. The maximum required signal-to-noise ratio, and thus the required number of bits per sample, is independent of the nominal sound pressure level.

3. The position of the 45 lines indicates the idle channel noise level, the is at the quiet hearing threshold. The idle channel noise level is related to the level of the Nominal sound pressure level Lso inversely proportional.

4. The inflection point of the straight line indicates the compander gain or the degree of companding. It depends on the bandwidth of channel i (1T, 2T or 3T) and the nominal sound pressure level Lso.

The information flow of a bandpass PCM signal is therefore exclusive determined by the width and number of bandpass channels. He is for B; <4T independent from the nominal sound pressure level.

The height of the nominal sound pressure level is limited solely by the Perceptibility of the empty channel noise in channel 1. The empty channel noise is reduced proportional to the increase in the degree of companding. By dimensioning the Compander characteristic is the maximum listening volume that is currently set still guaranteed freedom from noise.

Example 1 (see Fig. 8) Selected: Lso = 90 dB, B1 = 3T Then: (5ffl) PcM max = 65 dB, corresponding to 10.6 bit sample This results in compander gains Ki and degrees of companding Gj: fmx = 0.8 kHz: Kl = 31 dB G1 = 5.2 bit fm2 = 3.2 kHz : K2 = 39 dB G2 = 6.5 bit fm3 = 10 kHz: K3 = 20 dB G3 = 3.3 bit This means that the optimal quantizations for the 3 octave-wide channels of the bandpass PCM path fixed: fm, = 0.8 kHz: 15.7 to 10.6 bit / sample fm2 = 3.2 kHz: 17.0 to 10.6 bit / sample fm3 = 10 kHz: 13.8 to 10.6 bit / sample These bandpass channels are up to peak sound pressure levels of Ls = 100 dB noise-free.

If you were to build a corresponding octave band PCM route (8 octave band passes), so the resulting information flow would be I = # fsi. 10.6 bit = 340 kbit / s. Compared to a 16-bit linear PCM path (whose system dynamics are 6 dB smaller) this means a saving of 172 kbit / s (35%). Would you like that for 16-bit linear halve the required information flow (approx. 250 kbit / s), so the octave width would have to be the bandpass filters can be reduced by a factor of 0.74. For maintaining the system dynamics the degrees of companding would then have to be increased by a factor of 1.34.

But the specific gain per bit saving is only at high frequencies great. It is therefore cheaper to use a few narrow bandpass filters in the upper frequency range to be provided than wider bandpass filters in the entire frequency range. With the same profit the technical effort is then lower. However, a higher one is required Companding.

Example 2 Selected: Lso = 90 dB Bj = 1.5T (7.5 ... 10.7 and 10.7 ... 15 kHz) Then: (S »z) PCM; I, 1a,. = 35 dB, corresponding to 5.7 bit sample This results in an information flow of 348 kbit / s; the saving compared to 16-bit linear is 164 kbit / s. This is achieved by dividing the frequency band into the three bands 0.04 ... 7.5 / 7.5. . . 10.7 / 10.7 ... 15 kHz. The lower band works 16-bit linear. Compared to Example 1, practically the same information flow is achieved with three bands. These and other dimensioning examples are compiled in Table 2: Table 2 .lj Example if "; foj bit iErz ktiz sample kblt / s kbit / s 0.04 0.12 15.9 2.6 2 0.12 0.25 10.6 2.7 3 0.2 0.5 10.6 5.3 4 0.5 0.9 10.6 10.6 5 0.9 1.9 10.6 21.2 6 1.9 3.8 10.6 42.5 7 3.8 7.5 10.6 84.9 8 7.5 15.0 10.6 169.9 2 1 0.04 7.5 16.0 255.6 2 7.5 10.7 5.9 40.5 348.3 3 10.7 15.0 5.7 52.2 3 1 0.04 5.4 16.0 184.0 2 5.4 7.5 5.6 24.9 3016 3 7.5 10.7 5.9 40.5 4 10.7 15.0 5.7 52.2 4 1 0.04 5.0 16.0 170.4 2 5.0 10.0 10.6 113.2 354.2 3 10.0 15.0 6.6 70.6 5 1 0.04 3.8 16.0 129.5 2 3.8 7.5 10.6 84.9 384.3 3 7.5 15.0 10.6 169.9 6 1 0.04 6.0 16.0 204.5 2 6.0 9.0 6.6 42.4 350.4 3 9.0 15.0 8.1 103.5 7 1 0.04 6.2 16.0 211.3 2 6.2 7.7 4.1 12.8 3 7.7 9.6 4.1 15.5 286.8 4 9.6 12.0 4.1 21.0 5 12.0 15.0 4.1 26.2 The resulting information flows v li can be achieved solely by adapting the PCM path to the eavesdropping threshold. It can be seen that the division into three frequency bands compared to 16-bit linear can bring about a saving of about 150 kbit / s (Examples 2, 4, 6), and that the choice of the cut-off frequencies of the three bandpass filters is not solely based on the eavesdropping threshold is fixed.

Adaptation of the bandpass PCM channel to the amplitude statistics Therefore There is a dimensioning leeway with regard to the adaptation to the amplitude statistics. The additional savings that can be achieved are calculated using the static J17 pre / de-emphasis curve explained; it is correspondingly higher when using an adaptive adjustment the short-term spectrum through adaptive, complementary amplification.

Fig. 9 shows an embodiment for the static adaptation to the long-term spectrum. The J17 curve (dashed line) is shown in Compare this to the quiet hearing threshold. In the frequency range 0.5 ... 6 kHz (bandpass i = 1) the frequency curve of the pre / deemphasis is precisely matched to the resting hearing threshold (»Quiet hearing threshold emphasis«). In the range 6 ... 9 kHz (bandpass filter i = 2) a complementary gain of 12 dB, in the range 9 ... 15 kHz (bandpass filter i = 3) of 18 dB. Overall, the modulation range of this bandpass PCM path is in the frequency range 1 ... 9 kHz restricted by approx. 6 dB less than in the case of the J17 pre / deemphasis for broadband PCM links.

Nevertheless, the resulting information flow savings are still over 12.8 kbit / s higher: Bandpass 1: The quiet threshold distance is 12 dB. With2 bit / sample and a sampling frequency (0.04 ... 6 kHz) fs = 12.8 kHz is reduced the transfer rate by 25.6 kbit / s.

Bandpass 2: The attenuation of the noise level is 12 dB. With 2 bit / sample and fs = 6.4 kHz reduces (6 ... 9 kHz) the transmission rate is reduced by 12.8 kbit / s.

Bandpass filter 3: The attenuation of the noise level is 18 dB. With 3 bit / sampleundfs = 12.8 kHz reduces (9 ... 15 kHz) the transmission rate is reduced by 38.4 kbit / s.

This is a total of 76.8 kbit / s. With this band-pass PCM section (Example 6 in Table 2) the net information flow 274 kbit / s. Compared to a corresponding broadband PCM route (quantization 16 to 14, J17 pre / deemphasis) which have to transmit a net information flow of 447 kbit / s can pass through the division into just three bandpasses already save 173 kbit / s. This is in comparison for J17-Pre / Deemphasis the restriction of the modulation range in the frequency range 19 kHz lower by approx. 6 dB, and the sound quality of the two PCM systems is equivalent. In addition, the bandpass limitations cause the distortion products that occur due to clipping when overdriving, practical in the frequency range 6 ... 15 kHz be completely suppressed. Temporary overrides caused by the restriction of the modulation range according to FIG. 9, interfere with it Basically and as a result of the concealment practically not. In contrast to the broadband PCM route therefore, the use of the long-term amplitude statistics does not increase the headroom necessary.

Fig. 10 shows an embodiment for an additional dynamic Adaptation to the short-term spectrum.

An octave-wide frequency band division is shown according to Example 1 in Table 2. The information flow to be transmitted on this bandpass PCM link is 340 kbit / s without the use of the amplitude statistics. Simply inserting static, complementary preamplifications in the bandpass filters i = 7 and i = 8 of 6 or 18 dB (hatched areas and clip boundary curve in FIG. 10) would result in a saving of 8.0 kbiUs + 48.0 kbit / s = 56.0 kbit / s result (use of the long-term amplitude statistics). If one also provides an adaptive, complementary preamplification with a 12 dB limiter line in each of the bandpass filters i = 5 to i = 8 (dotted areas in Fig. 10), the information flow is reduced to 223.4 kbit / s: Table 3 #Ii example i f11 f0 bit l; i kHz Ez sample IcblUs kbit / s la 1 0.04 0.12 15.9 2.6 2 0.12 0.25 10.6 2.7 3 0.25 0.5 10.6 5.3 4 0.5 0.9 10.6 10.6 223.4 5 0.9 1.9 8.6 17.2 6 1.9 3.8 8.6 34.5 7 3.8 7.5 7.6 60.9 8 7.5 15.0 5.6 89.6 This bandpass PCM path has the following properties: 1) Overloads (exceeding the clip limit) in the frequency range 0.1 ... 15 kHz do not lead to annoying noises due to non-linear distortion products, but to far less annoying, brief sound colorations due to linear distortion. Overloads are less critical than with analog transmission links.

2) Lowering the clipping limit by 6 dB in the 3.8 frequency range. . . 7.5 kHz and by 18 dB in the range 7.5 ... 15 kHz (see Fig. 10) are therefore reduced Even with a critical program, the controllability of the track is not possible. The headroom can even be reduced without any problems.

3) Companding noise is imperceptible even at the highest monitoring levels. Only for pure tones with f> 1.9 kHz, the level of which is greater than -12 dB (referred to on the cliff limit) is theoretically that the eavesdropping threshold is exceeded the companding noise possible.

4) If by appropriate digital companding in all eight Bandpasses the bit rates (bit / sample) provided in Table 3 for the transmission are met, the companding noise is in the entire bandwidth 0.04 ... 15 kHz - regardless of the degree of companding - subliminal. Therefore can Any high dynamic range can be transmitted without audible losses. The maximal Dynamics are therefore only limited by the system dynamics of the A / D-D / A converter.

5) Due to the bandpass limitations and the complementary gains Disturbances caused by bit errors in the transmission are much less frequent and weaker as corresponding crackling disturbances or background noises in broadband PCM-Ub broadband PCM transmission.

Overall, for this embodiment example the invention results Bandpass PCM route compared to a corresponding broadband PCM route as follows Advantages: - The transmission bit rate for the bandpass PCM-coded AF signal is about 230 kbit / s.

- The maximum dynamic is only due to the resolution of the A / D-D / A converter limited.

- The companding noise is imperceptible regardless of the monitoring level.

- The route is less sensitive to overloading than analog ones Stretch.

- The route is insensitive to bit errors.

- blank page -

Claims (6)

Claims: 1. Method for the digital transmission of a radio program signal, in which the frequency band of the radio program signal is split into sub-bands is quantized differently as well as the individual subbands and then are transmitted in time division multiplex, characterized in that the quantization of each sub-volume through the following measures, either individually or in combination takes place: a) in accordance with the applicable level for a given level of ablation Quiet hearing threshold; b) in accordance with the requirements for the respective bandwidth of the sub-band applicable eavesdropping threshold; c) in accordance with the for the respective frequency position of the Subband applicable amplitude statistics of the broadcast program signal takes place. 2. The method according to claim 1, characterized in that the quantization is chosen for each sub-band such that the resulting quantization noise level approximately equal to the value of the quiet hearing threshold applicable for the respective subband is at the specified listening volume level. 3. The method according to claim 1 or 2, characterized in that each subband is companded and the characteristic curve for this companding is such it is chosen that the resulting quantization noise level is approximately the same the value present at the respective lower sub-band cut-off frequency for the respective upper subband limit frequency applicable monitoring threshold curve lies. 4. The method according to any one of claims 1 to 3, wherein the digitized Radio program signal is digital / analog converted on the receiver side, characterized in that that the subbands are preamplified differently before the quantization and after the digital / analog conversion at the receiver end, the differences in gain be balanced. 5. The method according to any one of claims 1 to 3, wherein the digitized Radio program signal is digital / analog converted on the receiver side, characterized in that that a sub-band before the quantization according to the complementary frequency curve the noise threshold is pre-distorted and that after the digital / analog conversion at the receiver end the pre-distortion is compensated. 6. The method according to claim 4 or 5, characterized in that the degree of pre-amplification and / or pre-distortion of the amplitude of the respective Subband is controlled. The invention is based on a method according to the preamble of Claim 1. One such method is from The Bell System Journal ", October 1976, pages 1069-1085. High quality broadcast program signals with frequency spectra up to 15 kHz is currently only digitally transmitted on cable and microwave links. It However, there are considerations about digital radio transmission in the VHF range in the future to be provided in order to achieve the quality of digital sound carriers also in radio. For the digital transmission of voice signals on PCM lines it is known ("The Bell System Technical Journal", October 1976, pages 1069 to 1085), the frequency band of the analog voice signal by a corresponding number of Split band passes into sub-bands with increasing bandwidth and each sub-band to quantize, the sampling frequency for each sub-band at least twice Bandwidth of the relevant subband corresponds. This results in different Sampling frequencies for the individual sub-bands. Furthermore takes place for the sub-bands a partially different quantization, so that a total of either specified transmission bit rate for the subsequent time division multiplex transmission the distribution of the quantization noise in the sense of a Can influence improvement of the signal quality or with a given signal quality can reduce the transmission bit rate. The object of the invention is that for the transmission Known methods provided by speech signals for the transmission of high-quality Broadcast program signals, especially in the VHF range, adapt to this Way a reduction of the transmission bit rate without deterioration of the signal quality to enable. This object is achieved according to the invention by the in the characterizing part of claim 1 specified features solved. Advantageous embodiments of the method according to the invention result from the subclaims. The invention is explained in more detail below with reference to FIGS. 1 to 10 explained. The minimum flow of information to be transmitted on a PCM link with a fixed sampling frequency depends only on the number of quantization levels that per sample must be transmitted at least on average in order to to maintain sufficient system dynamics. The system dynamics are sufficient if the quantization error (the quantization noise) of the PCM path is even is not audible. When reproducing, the sound pressure level of the quantization noise must be always below the quiet hearing threshold or the hearing threshold of the human hearing.