Analogue to digital converting

0001The invention concerns a method for converting an analog input signal into a digital output signal with at least two analog signal components, whose signal levels differ from each other, being derived from the input signal and the signal components being respectively subjected to an analog-digital conversion. The invention also concerns a device with which this method can be implemented.

0002There is a need for analog-digital transformation in many areas of digital signal processing, in particular in digital microphony, whereby signals with a dynamic range that is as wide as possible can be transmitted while simultaneously maximizing the signal-to-noise ratio (SNR). Qualitatively high performance A/D converters having a high resolution are used for this purpose. However, they are expensive and they use a lot of power, which is particularly disadvantageous in portable applications, e.g. radio microphones.

0003The state of the art attempts to circumvent these problems by splitting the analog input signal into two signal components, with the still analog signal components being subjected to different levels of amplification. The signal components present in different signal pathways now having different levels are independently subjected to an A/D conversion. If the level of the signal component, which already has a higher level in comparison with the other signal component because of the different amplification, increases above a threshold that causes clipping in the corresponding A/D converter, a switchover to the lower level signal component or the corresponding signal pathway occurs, in which, due to its lower level, no clipping takes place (as yet) in the A/D converter. This method is also called gain staging.

0004The advantage of this method is that at least two low-resolution converters can be used in place of one high-resolution converter without thereby limiting or influencing the dynamic range that is available for transmission. The major challenge in this method, however, is posed by switching between the two signal pathways. This requires an extremely precise matching of the digitized signal components to each other regarding both the amplitude and the phase relation between the signal components. Even time shifts in the nanosecond range can cause unwanted interferences at the "junctions" at which one signal component is attached to the other signal component. The effort expended by the complicated matching of the signal components is not related to the simplification of the actual A/D conversion.

0005The method described above is, for example, described in "<nplcit id="ncit0001" npl-type="s"><text>The Digitally Interfaced Microphone", S. Peus, O. Kern Neumann GmbH, Berlin, presented at the 110th AES Convention 2001 May 12-15, Amsterdam, The Netherlands</text></nplcit>. The disclosure of this publication as well as the disclosure in <patcit id="pcit0001" dnum="EP1253518B1"><text>EP 1 253 518 B1</text></patcit> is included in this description by reference.

0006Both of the aforementioned publications disclose another method for the A/D conversion of signals. This method too is based on replacing an expensive A/D converter having a high resolution with two A/D converters having a low resolution, which consume less power in combination. Contrary to the methods described above, signal components that are produced from the original analog input signal after passing through a nonlinear network are directed into the two signal pathways. One of the signal components is distorted so that it essentially contains no useful information from the input signal below a threshold signal level. The other signal component results from the addition of the input signal with correct sign to the distorted signal component (i.e. the distorted signal is subtracted from the input signal), whereby a complementary distorted signal shape is produced. After the A/D conversion of the two signal components, the reconstruction of the original signal occurs via the addition of the signal components, because the distortion of the signal components is always accurately complementary and thus resulting in an undistorted digital output signal. Because of the decrease in amplitude due to the subtraction of the distorted signal, the corresponding A/D converter does not reach the clipping level.

0007It is the object of the present invention to improve the initially described method and to thus relieve the problems of switching between the signal paths. Regarding the last method described above, disadvantages which occur due to the nonlinear network and its amplification, particularly with high signal and noise amplitudes, are relieved.

0008This object is achieved with a method of the type described above in that the digitized signal components are weighted with a variable weighting factor and the weighted signal components are added together.

0009The weighting and subsequent merging respectively addition of the signal components makes it possible to react to the changing characteristics of the analog input signal and thus also of the analog signal components in a simple manner. By means of the variable weighting factor, it is possible to smoothly crossfade the signal components. The fractional content of the signal components in the resulting output signal can thus be adapted to existing conditions.

0010It is basically possible to draw upon any determinable property of one of the two signal components or of the input signal as well as a criterion for a change in the weighting of the signal components, e.g. the signal magnitude, the signal-to-noise ratio, harmonic distortion, clipping, etc. The level of one of the two signal components has been found to be particularly advantageous and reliable as a reference parameter while the magnitude of the signal is nevertheless directly associated with the occurrence of clipping. Therefore, as soon as the magnitude of the signal exceeds a preset threshold value and clipping is expected to occur, the signal components are crossfaded, i.e. the weighting of the signal components in the resulting output signal changes. The crossfading occurs continuously, i.e. by a constant adjustment of the weighting factor, whereby the transition becomes smooth.

0011In a preferred embodiment, the change in the weighting factor is time-delayed. This is in particular done if the signal level decreases again after reaching a critical threshold value and if the weighting factor of the more amplified signal component is again to be represented more strongly in the output signal. Due to the time delay, i.e. a slow adjustment of the weighting, the crossfading is no longer perceptible to the listener.

0012In the following the invention is described in greater detail based on the drawings. <ul id="ul0001" list-style="none" compact="compact"><li><figref idref="f0001">Figure 1</figref> shows a schematic drawing of the generation and digitization of two signal components having different levels,</li><li><figref idref="f0002">Figure 2</figref> shows a block diagram of a circuit of this invention,</li><li><figref idref="f0003">Figure 3</figref> shows an extension of the idea of this invention to more than two signal pathways.</li></ul>

0013<figref idref="f0001">Figure 1</figref> shows in a schematic drawing that two signal components having different levels are conveyed into the signal paths 3, 3' from an analog input signal (audio in), which can, for example, be a signal from a microphone 1. In the example shown here, this occurs through differing amplifications by the amplifiers 2, 2', whereby a stronger amplification occurs in the signal pathway 3 than in the signal pathway 3'. It would also be possible to attenuate one of the two signal components, e.g. via a voltage divider.

0014The two analog signal components are preferably essentially shifted to the same phase and a constant magnitude ratio over the entire transmitted frequency range. This has a positive effect, in particular at low frequencies (&lt; 50-60 Hertz), in that the emergence of unwanted artifacts at low frequencies is prevented during the subsequent merging of the digitized signal components according to this invention.

0015The signal pathway 3, in which the signal component has the higher level, is hereafter also called the "more sensitive" signal pathway and the signal pathway 3', in which the signal component has a lower level, is also called the "robust" signal pathway. Both signal components are thereafter digitized in the respective A/D converters 4 and 4' and are subjected to further processing in a processing unit 5 and are merged so as to generate the output signal y.

0016<figref idref="f0002">Figure 2</figref> shows the internal workings of the processing unit 5 and thus the continuation of the two signal pathways 3, 3'. The level of the signal component x2 in the sensitive channel 3 is now already higher by a constant factor (e.g. between 10 and 40 dB) than that of the signal component x1 in the robust channel 3'. If the sensitive channel reaches clipping level, then crossfading to the other (robust) channel occurs. It is intended to achieve a higher ADC resolution and to create a system with an SNR of approximately 120 dB defined by the external wiring. The disadvantage of the bad SNR of only approximately 85 dB of the individual ADCs is circumvented by the smooth crossfading of the sensitive and the robust channel.

0017The process in the signal processor after the A/D transformation is described hereafter: each of the digitized signal components x11, x22 are weighted with a weighting factor bf or 1-bf and added in a weighting unit 6, also called a fader, whereby the output signal y of the device of this invention is formed: <maths id="math0001"><math display="block"><mi mathvariant="normal">y</mi><mo mathvariant="normal">=</mo><mi mathvariant="normal">x</mi><mo>⁢</mo><mn mathvariant="normal">22</mn><mo mathvariant="normal">*</mo><mfenced><mn mathvariant="normal">1</mn><mo mathvariant="normal">-</mo><mi>bf</mi></mfenced><mo mathvariant="normal">+</mo><mi mathvariant="normal">x</mi><mo>⁢</mo><mn mathvariant="normal">11</mn><mo mathvariant="normal">*</mo><mi>bf</mi><mo mathvariant="normal">,</mo></math><img file="EP2207264A1_D0001.tif" /></maths> where the weighting factor bf can have values between 0 and 1.

0018In the present example, the signal x1 is identical to x11, while x22 is obtained from the signal x2 by a correction in signal level. Before the two signal components are merged in the weighting unit 6, they must again have the same level in order to avoid unwanted variations in signal level in the output signal y when they are merged. For this purpose, x2 is appropriately attenuated with the attenuator dg. It is just as possible to amplify x1 instead, or the appropriate adjustment could take place in the fader with weighting factors modified accordingly.

Description of the appropriate algorithm:

0019x2 is the output signal of the ADC in the sensitive channel 3, x1 is the signal of the robust channel 3'. The signal x22 is produced from x2 by multiplication with a constant gain dg. The constant gain corresponds to the difference in signal level x1:x2. In the signal flow, a variable gain g, which has the function of adaptibly adjusting possible tolerances due to external component variations, follows the constant gain dg. A typical control range would be +/-1 dB. Since, in the present embodiment, the signal processing algorithm is performed in a fixed point computer, which does not permit multiplications with numbers greater than 1, this adaptive gain is achieved in two steps. The first step operates with a variable of about 0.5 and is followed by a fixed amplification by a factor of 2: <maths id="math0002"><math display="block"><mi mathvariant="normal">x</mi><mo>⁢</mo><mn mathvariant="normal">22</mn><mo mathvariant="normal">=</mo><mi mathvariant="normal">x</mi><mo>⁢</mo><mn mathvariant="normal">2</mn><mo mathvariant="normal">*</mo><mi>dg</mi><mo mathvariant="normal">*</mo><mi mathvariant="normal">g</mi><mo mathvariant="normal">*</mo><mn mathvariant="normal">2</mn></math><img file="EP2207264A1_D0002.tif" /></maths>

0020The weighting factors bf and 1-bf are calculated as a function of the level of one of the signal components. In the present example, the level of the signal component x2 in the sensitive channel 3 is used as the reference parameter. The magnitude abs(x2) can, for example, be determined in the form of the absolute value, the RMS (root mean square) of the peak value or the rectification value. Such level measurements are well-known to the specialist. An intermediate variable b is now calculated based on the signal x2: <maths id="math0003"><math display="block"><mi mathvariant="normal">b</mi><mo mathvariant="normal">=</mo><msup><mfenced><mi>abs</mi><mfenced><mi mathvariant="normal">x</mi><mo>⁢</mo><mn mathvariant="normal">2</mn></mfenced><mo mathvariant="normal">*</mo><mn mathvariant="normal">1</mn><mo mathvariant="normal">/</mo><mi>Thr_fade</mi></mfenced><mo mathvariant="normal">∧</mo></msup><mo>⁢</mo><mn mathvariant="normal">2</mn></math><img file="EP2207264A1_D0003.tif" /></maths><maths id="math0004"><math display="block"><mi>if</mi><mfenced><mi mathvariant="normal">b</mi><mo mathvariant="normal">></mo><mn mathvariant="normal">1</mn></mfenced><mo>⁢</mo><mi mathvariant="normal">b</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">1</mn></math><img file="EP2207264A1_D0004.tif" /></maths> where Thr_fade defines an adjustable threshold value, which essentially corresponds to the upper range of the A/D converter, but preferably lies somewhat below this range (e.g. about -6 dB), in order make it possible to crossfade in time to the robust channel 3', so that the occurrence of clipping in the ADC does not become audible in the output signal y.

0021If the level abs(x2) of the signal component x2 is greater than the threshold value Thr_fade, then b = 1, thereunder the intermediate variable b, follows a square characteristic, so that the intermediate parameter b decreases continuously quadratically with the falling signal magnitude abs(x2).

0022The weighting factor bf is produced from the intermediate variable b by a smoothing filter, with bf being rapidly adjusted with rising b, while it decreases according to a first order low-pass filter with falling b. This behavior is comparable with that of a release filter of an audio compressor. The smoothing filter can be described with the following algorithm. <maths id="math0005"><math display="block"><mi>bf</mi><mo>=</mo><mi>bf</mi><mo>*</mo><mi>Coeff</mi><mo>;</mo></math><img file="EP2207264A1_D0005.tif" /></maths><maths id="math0006"><math display="block"><mi>If</mi><mspace width="1em" /><mfenced><mi mathvariant="normal">b</mi><mo mathvariant="normal">></mo><mi>bf</mi></mfenced><mspace width="1em" /><mi>bf</mi><mo mathvariant="normal">=</mo><mi mathvariant="normal">b</mi><mo mathvariant="normal">;</mo></math><img file="EP2207264A1_D0006.tif" /></maths> where <maths id="math0007"><math display="block"><mi>coeff</mi><mo mathvariant="normal">=</mo><msup><mi mathvariant="normal">e</mi><mo mathvariant="normal">∧</mo></msup><mo>⁢</mo><mfenced><mo mathvariant="normal">-</mo><mn mathvariant="normal">2.2</mn><mo mathvariant="normal">/</mo><mfenced><mi>SampleRate</mi><mo mathvariant="normal">*</mo><mi>TimeConst</mi></mfenced></mfenced></math><img file="EP2207264A1_D0007.tif" /></maths>

0023SampleRate is the sampling rate of the ADC and the time response is described by the constant TimeConst. TimeConst is typically selected at 0.1 seconds, from which a bf value of 0.9995 is obtained.

0024The algorithm described above represents a possible implementation of the invention. The algorithm has the following advantages: because of the square dependence of the weighting factors bf and respectively 1-bf on the signal magnitude abs(x2), the system reacts very rapidly to increasing signal level and the weighting of the signal components in the resulting output signal is changed rapidly. If the signal magnitude abs(x2) decreases again, the weighting factors bf or respectively 1-bf are updated with a time delay. This inertia of the system with decreasing signal strength ensures a smooth crossfading of the two signal paths and ensures that the crossfading process does not become audible to the user. The selection of the time constants TimeConst of approximately 0.1 s is intended to execute the crossfading within the masking interval of human auditory perception.

0025The weighting unit 6 (the fader) is then controlled with the weighting factors bf or respectively 1-bf. In the limiting case, if bf = 1, only the signal component x11 of the robust channel 3' arrives at the output y. In the other limiting case, if bf = 0, only the signal component x22 of the sensitive channel 3 arrives at the output y. Smooth crossfading occurs in between.

0026In a preferred embodiment, a provision for the adaption of a previously mentioned gain g is made, which has the function of adaptively adjusting possible tolerances due to component variations:

0027The average value MeanX11 or MeanX22 of the respective absolute values abs(x11) or abs(x22) is determined from the signals x11 and x22. The two average values are subtracted from each other resulting in the difference delta which is used to modify the Gain g(n) (closed loop), where: <maths id="math0008"><math display="block"><mi>delta</mi><mo>=</mo><mi>MeanX</mi><mo>⁢</mo><mn>11</mn><mo>-</mo><mi>MeanX</mi><mo>⁢</mo><mn>22</mn></math><img file="EP2207264A1_D0008.tif" /></maths><maths id="math0009"><math display="block"><mi mathvariant="normal">g</mi><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">=</mo><mi mathvariant="normal">g</mi><mo>⁢</mo><mfenced><mi mathvariant="normal">n</mi><mo mathvariant="normal">-</mo><mn mathvariant="normal">1</mn></mfenced><mo mathvariant="normal">+</mo><mi>delta</mi></math><img file="EP2207264A1_D0009.tif" /></maths> n is the corresponding sampled value.

0028To avoid side effects and misadaptions, the gain g(n) is only adjusted if the weighting factor bf is within a range of, for example, 0.0001 to 0.9.

0029Because of the A/D conversion with a non-ideal ADC the signals contain a DC component. In order to exclude malfunctions from subsequent parts of the algorithm, this component is removed on the input side by means of an appropriate filter, a DC bias filter, which behaves like a high-pass filter.

0030As mentioned previously, certain actions are preferred in the analog part of the circuit to make it possible to also particularly transmit low frequencies better via the method of this invention. To guarantee the correct functioning of the following algorithm, it is important for the two analog signal components to have a level ratio that is as constant as possible over all frequencies as well as no phase shift. Phase shifts and frequency dependent differences in signal level act as distorting factors at very low audio frequencies since the fader follows the signal shape of the audio signal directly in this case.

0031The invention is naturally not limited to the embodiment that is shown. It is in particular possible to build one or more latency components into the signal paths. This enables a somewhat time-delayed increase of the weighting factor bf, so that a smooth crossfading of the signal paths is enabled before clipping occurs.

0032It is also possible for other algorithms to act as the basis of the generation of the weighting factors as a function of one of the signal components. In particular, the square dependency of the weighting factors on the signal level is not mandatory. The time delay in adjusting the weighting factor can also behave differently than a 1^st order low-pass filter.

0033As an extension of this invention, a further variant shown in <figref idref="f0003">Figure 3</figref> contains several signal paths 3, 3', 3", 3"'. In this example, more than two signal components are derived from the analog input signal, with each signal component having a different signal level. Starting from channel 3, also called the "most sensitive channel", having the highest signal level, it is thus possible to successively crossfade to channels with respectively lower signal strengths. <figref idref="f0003">Figure 3</figref> shows a possible way to implement this idea, with the signal paths 3, 3', 3", 3"' and the corresponding weighting units 6, 6', 6" being nested so as to respectively merge two signal paths. If the most sensitive channel 3 reaches the clipping level, blending over to the next channel 3' having a lower signal strength occurs. If the latter also reaches the clipping level, crossfading to channel 3" having the next-lower signal strength occurs, and so on.