TW201532035A - Prediction-based FM stereo radio noise reduction - Google Patents

Abstract

The present document relates to audio signal processing, in particular to an apparatus and a corresponding method for improving the audio signal of an FM stereo radio receiver. In particular, the present document relates to a method and system for reducing the noise of a received FM stereo radio signal. An apparatus (2) configured to reduce noise of a received multi-channel FM radio signal is described, wherein the received multi-channel FM radio signal is representable as a received mid signal and a received side signal. The apparatus (2) comprises a parameter determination unit (77) configured to determine one or more parameters indicative of a correlation and/or decorrelation between the received mid signal and the received side signal; and a noise reduction unit (79) configured to generate a noise-reduced side signal from the received mid signal and not from the received side signal using the one or more parameters.

Description

Predictive FM stereo radio noise reduction

This document relates to audio signal processing, and more particularly to apparatus and corresponding methods for improving the audio signal of an FM stereo radio receiver. In particular, this document relates to methods and systems for reducing noise in received FM stereo radio signals.

In an analog FM (frequency modulation) stereo radio system, the left channel (L) and the right channel (R) of the radio signal are conveyed in the middle (M/S) representation, that is, the center channel (M). With side channel (S). The middle channel M corresponds to the sum (sum) signal of L and R, for example, M=(L+R)/2, and the side channel S corresponds to a difference signal of L and R, for example, S=(LR)/2 . For transmission, the side channel S is modulated up to a 38 kHz rejection carrier and added to the baseband signal M to form a backwards-compatible stereo multiplex signal. This multiplexed baseband signal is then used to modulate the HF (high frequency) carrier of the FM transmitter, typically operating in the range between 87.5 and 108 MHz.

When the reception quality is reduced (ie, on the radio channel) The signal on the signal-to-noise radio is reduced. The S channel is usually worse than the M channel during transmission. In many FM receiver implementations, the S channel is muted when the reception conditions become too noisy/too noisy. This means that the receiver falls back from stereo to mono (mono) in case of poor HF radio signals (usually referred to as mono dropout).

Even if the medium signal M can accept quality, the side signal S can be noisy, so when mixing in the left and right channels of the output signal (for example, according to L = M + S and R = When MS is derived), it is possible to severely reduce the quality of the entire street. When the side signal S only has a bad quality for the intermediate quality, there are two options: the receiver does not choose to receive the noise related to the side signal S and outputs the real stereo signal containing the noisy/noisy left and right signals. That is, the receiver drops the side signal S and falls back to mono.

Parametric Stereo (PS) is a technique from the field of very low bit rate audio writing. The PS allows the 2-channel stereo audio signal to be encoded as a mono downmix signal combined with additional PS side information, ie PS parameters. Obtain a mono downmix signal as a combination of two channels of stereo signals. The PS parameter enables the PS decoder to reconstruct the stereo signal from the mono downmix signal and the PS side signal. In general, the PS parameters are time-and-frequency-variables, and PD processing in the PS decoder is typically implemented in a hybrid filter band plus a plurality of Quadrature Mirror Filter (QMF) rows.

The PS encoding using the received FM stereo signal has been proposed in WO 2011/029570, PCT/EP2011/064077 and PCT/EP2011/064084 in order to reduce the noise contained in the received FM stereo signal. The general principle of Parametric Stereo based FM stereo radio noise reduction technology is to use the parametric stereo parameters derived from the received FM stereo signal in order to reduce the left and right signals included in the reception. The noise. The disclosure of the above patent documents is incorporated by reference.

In this document, a method and system for FM stereo radio noise reduction using a prediction-based framework is illustrated. This predictive framework is an alternative to the parametric stereo (PS) based framework as indicated above. As will be explained in this document, predictive frameworks provide lower computational complexity. Further, it has been observed that the simultaneous predictive FM stereo radio noise reduction scheme achieves improved audio quality compared to the PS-based FM stereo radio noise reduction scheme.

According to one aspect, a device or system configured to reduce noise of a received multi-channel FM radio signal is illustrated. The multi-channel FM radio signal can be a two-channel stereo signal. In particular, the received multi-channel FM radio signal may be represented or may be presented as or representing a medium signal and a side signal. Further, the side signal can indicate the difference between the left signal and the right signal of the stereo signal.

In an embodiment, the device includes a parameter decision unit configured to determine one or more parameters indicative of correlation and/or decorrelation between the received intermediate signal and the received side signal. The one or more parameters may be the prediction parameter a for determining the correlation component of the noise reduction side signal from the received medium signal, and/or may be the correlation parameter b for determining the noise reduction from the decoupling form of the medium signal. The relevant components of the side signal. Further, the device includes a noise reduction unit configured to generate a noise reduction side signal from the received medium signal using one or more parameters. For this purpose, the noise reduction unit does not take into account the received side signals (eg, samples of the received side signals). In other words, the received side signal is not in the signal path used for the decision of the noise reduction side signal. In particular, the noise reduction unit can be configured to determine the noise reduction side signal only from the received medium signal (eg, the received medium signal sample) and one or more parameters.

As indicated above, the parameter decision unit can be configured to determine the prediction parameter a. The prediction parameter a may indicate a cross-correlation between the received intermediate signal and the received side signal. In particular, the parameter decision unit can be configured to determine the prediction parameter a based on the expected value of the product of the received medium signal and the corresponding sample of the received side signal. Even more particularly, the parameter decision unit can be configured to determine the prediction parameter a using the formula a=E[S*M]/E[M*M], where E[. ] means the desired operand, S means the received side signal and M means the received medium signal.

In the case where the parameter decision unit provides the prediction parameter a, the noise reduction unit can be configured to use the prediction parameter a to generate from the received signal. The noise is reduced by the side signal (or the noise is reduced by the side signal). The noise side low signal pair correlation component can be determined as the product of the prediction parameter a and the received medium signal, that is, a*M. This means that the relevant component of the noise reduction side signal can be a weighted form of the received medium signal. In view of the fact that the prediction parameter a can be a time variable and/or a frequency variable, the weighting factor for the received medium signal can be a time variable and/or a frequency variable.

The parameter decision unit can be configured to determine a decorrelation parameter b indicative of a decorrelation between the received intermediate signal and the received side signal. In particular, the parameter decision unit can be configured to determine the decorrelation parameter b based on the received side signal and the energy of the difference signal of the signal determined from the medium signal using the prediction parameter a. Even more particularly, the parameter decision unit can be configured to use the formula b=sqrt(E[D*D]/E[M*M]) with D=Sa*M as the difference signal to determine the decorrelation parameters b. The operand "sqrt()" indicates the square root operation.

In this case, the noise reduction unit can be configured to use the decorrelation parameter b to generate a noise reduction side signal (or a decorrelated component of the noise reduction side signal) from the decorrelated form of the received medium signal. In particular, the decorrelation component of the noise reduction side signal can be determined as b*decorr(M), with decorr(M) as the decoupling form of the received medium signal. The de-correlated form of the received medium signal can be determined by filtering the received medium signal by using an all-pass filter.

If the received side signal contains a large amount of noise, it can advantageously reduce the influence of the decorrelated component of the noise reduction side signal on the noise reduction side signal. For this purpose, the parameter decision unit can be configured to determine the receiving side The impact factor characteristic of the spectral flatness of the signal (or the spectral flatness of the received side signal). High spectral flatness typically indicates a high degree of noise contained within the side signal. As such, the decorrelation parameter b can depend on the impact factor. In particular, when the influence factor indicates the degree of increase in the spectral flatness of the received signal, the decorrelation parameter b can be reduced. By way of example, the impact factor is the SMF_impact_factor described in this document and the modified decorrelation parameter b_new is determined to be b_new=(1-SMF_impact_factor)*b, thereby forcing the decorrelation component of the low-side signal of the noise ( That is, b_new*decorr(M)) to zero, if the SMF-impact_factor tends to face "1".

As indicated above, the parameter decision unit can be configured to determine one or more parameters (eg, prediction parameter a and/or decorrelation parameter b) in a time variable manner. As such, for each of one or more parameters, a sequence of separate parameters for the corresponding sequence of time intervals can be determined. By way of example, for a first parameter (eg, prediction parameter a or decorrelated parameter b), a sequence of first parameters for a sequence of time intervals is determined. The sequence of time intervals can be a sequence of signal time frames (which include, for example, 2048 signal samples). In general, the sampling of the received intermediate signal and/or the received side signal (which is within a particular time interval) is used to determine a particular first parameter in the sequence of first parameters for a particular time interval in the sequence of time intervals. . In the case where one or more of the parameters are time variables, the noise reduction unit can be configured to generate a noise reduction side signal using one or more time variable parameters.

In order to ensure continuity between adjacent time intervals, and to avoid hearing discontinuities at the boundaries of adjacent time intervals, it may be advantageous to interpolate adjacent first parameters by sequence from the first parameter To determine the sequence of the first parameter of the interpolation.

In the case of highly degraded reception conditions, the FM receiver can force the received FM radio signal to mono, ie the FM receiver can suppress the received side signal. The device can be configured to detect such mono dropouts, ie the device can be configured to detect that the received multi-channel FM radio signal is a forced mono signal. This can be achieved by detecting a fast transition of the side signal from high energy to low energy reception. In particular, the energy of the side signal received during the first time interval in the sequence of time intervals can be determined and can be determined to be higher than the high threshold. Further, a number of transition periods following a continuous time interval may be determined during which the energy of the side signal falls from above the high threshold to below the low threshold. Based on this information, if the number of consecutive time intervals of the transition period is lower than the interval threshold, it can be determined that the received multi-channel FM radio signal following the first time interval is a forced mono signal. This interval threshold can be 1, 2, 3 or 4 time intervals following the first time interval.

If it is detected that the multi-channel FM radio signal received in the time interval (directly) following the first time interval is a forced mono signal, the parameter decision unit can be configured to be used for the first time interval. One or more parameters determine one or more parameters for (directly) following the time interval of the first time interval. In other words, the parameter decision unit can be configured to be concealed in mono down by using one or more parameters determined before the mono drop The lack of parameters during the withdrawal period.

As described above, the parameter decision unit can be configured to determine one or more parameters (eg, prediction parameter a and/or decorrelation parameter b) in a frequency variable manner. This means that different parameters are determined for the different sub-bands of the received medium and/or side signals. To this end, the device may include a mid-conversion unit configured to generate a plurality of mid subband signals covering the corresponding plurality of frequency ranges from the received intermediate signal. Further, the device can include a side transform unit configured to generate a plurality of side subband signals covering the plurality of frequency ranges from the received side signals. In such cases, the parameter decision unit can be configured to determine one or more parameters for each of the plurality of frequency ranges. In particular, for a second one or more parameters (eg, prediction parameter a and/or decorrelation parameter b), the plurality of second sub-band parameters may correspond to a plurality of neutron band signals and corresponding plurality of side parameters Take the signal to decide. This can be done by applying the above formula for determining one or more parameters (eg, prediction parameter a or decorrelation parameter b) to each of a plurality of frequency ranges.

The noise reduction unit can be configured to generate noise reduction side signals using one or more frequency variable parameters. In particular, the noise reduction unit can be configured to generate (a) only a plurality of noise reduction side subband signals from the corresponding plurality of neutron band signals and corresponding plurality of subband parameters. Using the inverse transformation unit, the side signal of the noise reduction can be generated by reducing the side subband signals from a plurality of noises.

The medium transform unit and/or the side transform unit may be a QMF filter bank and the inverse transform unit may be an inverse QMF filter bank. In view of the fact that the received medium signal is in the signal path (and the received side signal is not in the signal path), the side transform unit can meet lower requirements than the medium transform unit in terms of at least one of the following items. : frequency selectivity; frequency resolution; time resolution; and numerical accuracy.

The received FM radio signal can be dominated by the noisy received side signal, which has a higher energy level than the received medium signal. Such situations can lead to perceptually annoying artifacts when one or more parameters are used to generate noise reduction side signals from the received intermediate signal. To cope with such situations, the parameter decision unit can be configured to limit one or more parameters by applying a limit factor c to one or more parameters a. In particular, one or more parameters can be divided by a limiting factor c. In an embodiment, for c > 1, the limiting factor c is proportional to the sum of one or more squared parameters. In another embodiment, for c > 1, the limiting factor c is proportional to the square root of the sum of one or more squared parameters. In general, the selection of the limit factor c is such that the application of the limit factor c does not increase one or more parameters.

It should be noted that the device may include a delay unit configured to delay the received intermediate signal (sampling) by an amount of time corresponding to the computation time required to generate the noise reduction side signal (the corresponding sample).

In a good reception condition when the received side signal contains almost no noise, it is advantageous to use the received side signal for generating a stereo signal. For this purpose, the device can comprise a combined unit configured for use A quality indicator indicating the quality of the received multi-channel FM radio signal from the noise reduction stereo signal and the received side signal determines the modified noise reduction side signal. Depending on the quality of the received side signal, the modified noise reduction side signal can be between the noise reduction side signal and the received side signal (or selected from the noise reduction side signal and the received side signal or inserted in the noise reduction). The side signal is mixed with the received side signal. To this end, the combining unit may include a noise reduction gain unit configured to use noise reduction gain to reduce noise side signal weighting; a bypass gain unit configured to weight the received side signal using a bypass gain; The merging unit is configured to combine (eg, add) the weighted noise reduction side signal and the weighted received side signal; wherein the noise reduction gain and the bypass gain are dependent on the quality indicator. It should be noted that the combining unit can be configured to determine the modified noise reduction side signal in a frequency selective manner.

The device may include a quality decision unit configured to determine a quality indicator indicative of the quality of the received side signal. This can be determined by the power of the received medium signal (referred to as the medium power) and the power of the received side signal (referred to as the side power). The ratio of the medium power to the side power, that is, the mid-to-side ratio, can be determined, and the quality indicator of the received FM radio signal can be determined based on at least the mid-to-side ratio. This document describes various embodiments for determining the quality indicator α _HQ indicative of the received side signal quality in a trusted manner.

The device may further include an MS-to-LR converter configured to determine noise from the received middle signal and noise reduction side signal (or modified noise reduction side signal) to reduce the left signal and the noise to reduce the right signal. especially, The MS-to-LR converter can be configured to reduce the left signal from the sum of the received medium signal and the (modified) noise reduction side signal; and from the received medium signal and (modified) noise reduction side The difference between the signals determines the noise to reduce the right signal.

According to another aspect, a method for reducing noise of a received multi-channel FM radio signal is illustrated. The received multi-channel FM radio signal may be presented as a received medium signal and a received side signal. The method can include determining, between the received intermediate signal and the received side signal, one or more parameters indicative of correlation and/or decorrelation; and using the one or more parameters from the received intermediate signal instead of the received side signal Generate noise to reduce the side signal.

According to another aspect, the software program is described. The software program can be adapted to execute on a processor and, when implemented on a computing device, to perform the method steps outlined in this document.

According to another aspect, the storage medium is illustrated. The storage medium may include a software program adapted to be executed on the processor and, when implemented on a computing device, for performing the method steps outlined in this document.

According to another aspect, a computer program product is described. The computer program can include executable instructions adapted to be executed on the processor and, when implemented on a computer, for performing the method steps outlined in this document.

It should be noted that the methods and systems including their preferred embodiments as outlined in this patent application can be used independently or in combination with other methods and systems disclosed in this document. Further, all aspects of the methods and systems outlined in this patent application can be combined at will. In particular, the features of the scope of the patent application can be combined with each other in an arbitrary manner.

1‧‧‧ Receiver

2‧‧‧ Equipment

3‧‧‧PS parameter estimation unit

4‧‧‧Upmixing unit

5‧‧‧PS parameters

7‧‧‧PS encoder

8‧‧‧ part of the PS decoder

9‧‧‧Shrink mixing unit

10‧‧ ‧Resolver

20‧‧‧Detection unit

30‧‧‧Bypass gain unit

31‧‧‧ Noise reduction gain unit

32‧‧‧Merge unit

71‧‧‧Filter row

72‧‧‧ Filter row

73‧‧‧Filter row

74‧‧‧Delay

75‧‧‧LR-to-MS converter

76‧‧‧MS-to-LR converter

77‧‧‧Parameter decision unit

78‧‧‧Resolver

79‧‧‧ Noise Reduction Unit

The invention is explained by way of example with reference to the accompanying drawings, in which FIG. 1 illustrates a schematic example of a system for improving the stereo output of an FM stereo radio receiver; FIG. 2 illustrates an audio based on the concept of parametric stereo Example of a processing device; Figure 3 illustrates an example of an audio processing device based on the concept of prediction; Figure 4 illustrates an example power spectrum for a noisy FM radio voice signal and side signals; Figure 5 illustrates the use of a received FM radio signal A method flow diagram for processing a received FM radio signal; and FIG. 6 illustrates an example state machine for concealing prediction and decorrelation parameters.

FIG. 1 illustrates a schematic example system for improving the stereo output of an FM stereo radio receiver 1. As discussed in the prior art paragraph of this case, the stereo signal is transmitted in the FM radio as the design of the medium signal M and the side signal S. In the FM receiver 1, the side signal is used to establish a stereo difference between the left signal L and the right signal R at the output of the FM receiver 1 (at least when the reception is good enough and the side signal information is unmasked). In other words, the side signal is used to establish left and right audio signals from the middle signal. The left and right signals L and R can be digital or analog signals.

In order to improve the left and right audio signals L, R of the FM receiver, an audio processing device 2 that produces stereo audio signals L' and R' at its output can be used. The parametric stereo can be used to enable the audio processing device 2 to perform the noise reduction of the received FM wireless energy signal. Alternatively, the noise reduction of the received FM radio signal can be performed using the predictive parametric enablement audio processing device 2 as described in this document.

The audio processing in the device 2 is preferably implemented in the digital domain; therefore, in the case of an analog interface between the FM receiver 1 and the audio processing device 2, it is used before the digital audio processing in the device 2 Analog-to-digital converter. The FM receiver 1 and the audio processing device 2 may be integrated on the same semiconductor wafer or may be part of two semiconductor wafers. The FM receiver 1 and the audio processing device 2 can be part of a wireless communication device, such as a cellular telephone, a personal digital assistant (PDA), and a smart phone. In this case, the FM receiver 1 can be part of a baseband chip with additional FM radio receiver functionality. In another application, the FM receiver 1 and the audio processing device 2 can be part of a vehicle audio system to compensate for variations in receiving conditions for the moving vehicle.

Instead of using the left/right representation at the output of the FM receiver 1 and the input of the device 2, a mid/side representation can be used at the interface between the FM receiver 1 and the device 2 (see M, S in Figure 1). Used for the mid/side representation and L, R for the left/right representation). At FM receiver 1 and device 2 Such a mid/side representation at the interface may result in a reduced processing load since the FM receiver 1 has received the mid/side signal and the audio processing device 2 can directly process the mid/side signal without downmixing or L/ R to M/S conversion. If the FM receiver 1 is tightly integrated with the audio processing device 2, especially if the FM receiver 1 and the audio processing device 2 are integrated on the same device (for example, the same semiconductor wafer), the mid/side representation may be advantageous. of.

Alternatively, the radio signal strength signal 6 indicates that the radio reception condition is available for adaptation to audio processing in the audio processing device 2.

The combination of the FM radio receiver 1 and the audio processing device 2 corresponds to an FM radio receiver with an integrated noise reduction system.

2 illustrates an embodiment of an audio processing device 2 based on parametric stereo. The device 2 comprises a PS parameter estimation unit 3. The parameter estimation unit 3 is configured to determine the PS parameter 5 (which may or may not be left/right or medium/side representative) based on the input audio signal to be improved. PS parameter 5 may include (and other parameters) parameters indicating inter-channel intensity differences (IID or also CLD-channel level differences) and/or indicating inter-channel cross-correlation (ICC; Inter-channel cross-correlation) parameters. Preferably, PS parameter 5 is a time and frequency variable. In the case of the M/S representation at the input of the parameter estimation unit 3, however, by applying an appropriate conversion of the L/R channel, the parameter estimation unit 3 can decide the PS parameter 5 for the L/R channel.

The downmix audio signal DM is derived from the input signal. In case the input audio signal has been used in the middle/side representation, the downmix audio signal DM can be directly Corresponding to the middle signal. In case the input audio signal has a left/right representation, the audio signal can be generated by downmixing the audio signal in the downmix generating unit 9. Preferably, the result signal after the downmixing satisfies the middle signal M and can be generated by the following equation: DM=(L+R)/d, for example, having d=2, that is, the downmix signal DM satisfies the average of L and R. For different values of the scaling factor d, the average of the L and R signals is amplified or attenuated. The downmix generating unit 9 and the parameter estimating unit 3 are part of the PS encoder 7.

The device further includes an upmix unit 4, which is also referred to as a stereo mixing module or a stereo upmixer. The upmixing unit 4 is configured to generate stereo signals L', R' based on the audio signal DM and the PS parameter 5. Preferably, the upmixing unit 4 uses not only the DM signal but also the side signal S _o (which usually corresponds to the original received side signal S) or the pseudo side signal generated by the decorrelator 10 from the downmix signal DM (pseudo side signal). )S*. The decorrelator 10 receives the mono downmix DM and generates a decorrelated signal S*, which is used as a false side signal. The decorrelator 10 is as described in the document "Low Complexity Parametric Stereo Coding in MPEG-4", Heiko Purnhagen, Proc. Digital Audio Effects Workshop (DAFx), pp. 163-168, Naples, IT, Oct. 2004. Implemented by a suitable all-pass filter as discussed in Section 4. The discussion of its parametric parameters, in particular regarding the decision of the parametric stereo parameters, and in particular in section 4, is hereby incorporated by reference. The stereo mixing matrix 4 can be a 2x2 liter mixing matrix that produces stereo signals L', R' from the signals DM and S _o or S*. The upmixing unit 4 and the decorrelator 10 are part of the PS decoder 8.

Device 2 is based on the idea that the received side signal may be too noisy to reconstruct the stereo signal by simply combining the received side and side signals; however, in this case, the side signal received in the received L/R signal The component of the side signal or the signal may still be sufficiently good for stereo parameter analysis in the PS parameter estimation unit 3. The resulting PS parameter 5 can then be used to generate stereo signals L', R' which have a reduced number of levels of noise compared to the audio signal directly at the output of the FM receiver 1.

Therefore, noisy radio signals can be "cleaned up" by using the parametric stereo concept. Distortion and the main part of the noise in the FM radio signal are tied in the side channel, which is usually not used in PS downmixing. However, even in the case of noisy reception, the received side channel S typically has sufficient quality for PS parameter extraction.

In the drawing shown in this document, the input signal to the audio processing device 2 is a left/right stereo signal. With minor modifications to some of the modules within the audio processing device 2, the audio processing device 2 can also process input signals in the mid/side representation. Thus, the concepts discussed herein, as well as the input signals in the mid/side representation, can be used.

The PS-based FM stereo noise reduction method illustrated in FIG. 2 is well implemented for the side signals of the FM radio signals received therein to contain high or intermediate impurities originating from the radio transmission channel. The situation of the number of levels of the news. However, PS-based FM stereo The noise reduction method has several drawbacks. The PS-based FM stereo noise reduction method is quite computationally complex, when it requires two QMF analysis rows (for the calculation of PS parameters) and two QMF synthesis rows (for noise reduction stereo signals L', R' When produced). Further, PS-based FM stereo noise reduction methods typically utilize hybrid (ie, QMF plus additional Nyquist) filter bank techniques for adding at lower frequencies. Frequency resolution. This means that the decision of the PS parameter usually requires a high amount of filter bank operations. In addition, PS-based noise reduction methods require transcendental computations, such as sin() and atan() operations, which involve high computational complexity. Another disadvantage of the PS-based FM stereo noise reduction method is that it is not completely mono compatible, since it not only modifies the side signals but also modifies the medium signals to reduce the stereo signals L', R' in order to determine the noise. In other words, the mono downmix M' = (L' + R')/2 of the output of the PS-based FM stereo noise reduction system is typically different from the original signal M. In particular, if the received stereo signal has a wide stereo image (ie, if the received stereo signal has significant pan and/or decorrelated signal components), the mono downmix signal M' is typically Attenuated (ie, lower order). In contrast, for a predictive FM stereo noise reduction system, the mono downmix of the output is the original medium signal (since only the side signal is modified/processed).

The computational complexity of the PS-based FM stereo noise reduction method is a concern in many implementations, and this document illustrates an alternative framework for FM stereo noise reduction using predictive techniques. phase Predictive frameworks require lower computational complexity than parametric stereo (PS)-based frameworks. In particular, the predictive FM stereo noise reduction method uses a reduced number of filter banks and avoids using overrun calculations. At the same time, it has been shown that improved audio quality can be achieved when using the predictive FM stereo noise reduction method.

As outlined above, the PS-based FM radio noise reduction system illustrated in Figure 2 requires two QMF analysis filter banks and two QMF synthesis filter banks. All of these rank operations are in the signal path and therefore require high precision. The two QMF analysis filters are arranged at the outputs of the PS encoder 7 on signals L and R, and the two QMF synthesis filters are arranged at the output of the PS decoder 8 to produce signals L' and R'. Further, the PS-based system uses stereo parameters IID and ICC, and requires a transcendental function like sin() and atan() to calculate the elements of the stereo upmix matrix 4 from these stereo parameters.

It is proposed to reduce the computational complexity of the FM stereo noise reduction system by using a predictive framework instead of the downmix/upmix frame of the PS-based system described in FIG. By using the LR-to-MS converter 75 and the MS-to-LR converter 76 to switch to the mid/side signal representative, it is possible to reduce the number of required QMF rows in combination with the predictive method. The LR-to-RS converter 75 generates the middle signal M=(L+R)/2 and the side signal S=(LR)/2, and it can be omitted if the middle/side signal from the FM receiver 1 is directly The feed enters the audio processing device 2 of FIG. The MS-to-LR converter 76 performs an inverse operation on the LR-to-MS 75.

Figure 3 depicts an exemplary predictive FM radio noise reduction system An overview of the system, in which the thin line 80 means the time domain signal, the thick line 81 means the QMF domain signal, and the dotted line 82 means the parameter. The predictive framework uses only the QMF analysis filter bank 71 and a QMF synthesis filter bank 72 in the signal path, and a second QMF analysis bank 73 that is only used for parameter estimation (and which typically has reduced accuracy requirements).

As outlined above, PS-based FM radio noise reduction systems typically use a hybrid filter bank (i.e., a QMF row with an additional band split for the lowest QFM band using a Nyquist filter bank) for Higher frequency resolution is achieved up to a minimum frequency of approximately 1 kHz. For predictive FM radio noise reduction systems, it has been found that good audio quality can be achieved even without additional band-splitting by the hybrid filter bank. From then on, the predictive FM radio noise reduction system can use only the QMF row (ie, no hybrid filter bank), which further reduces computational complexity and also reduces the delay (or latency) of the calculation of FM radio signal processing. .

The predictive FM noise reduction system of Figure 3 is intended to generate a noise reduction side signal S' from the received intermediate signal M using two parameters a and b. The received medium signal M remains unchanged (the delay 74 is compensated for the calculation time required to determine the noise reduction side signal S'). This is a PS-based FM noise reduction system in which two signals, noise reduction left and right signals L', R' are determined as a function of the PS parameters.

The medium and side signals M and S defining the reception are M=(L+R)/2 and S=(LR)/2, and the side signal can use the prediction coefficient a and the residual signal D to represent S= a *M+D. This means that the prediction parameter a is used to predict the side signal from the medium signal. The best prediction coefficient a (which minimizes the energy of D) can be calculated as a = E[S*M]/E[M*M], where E[. ] means the expected operand. Literally, the prediction coefficient a can be determined as the ratio of the cross-correlation between the receiving side and the received medium signal to the energy of the medium signal. In general, the coefficients a (and b) are time and / or frequency variables. This means that different coefficients a (and b) are used for different time intervals and/or different frequency ranges. As a result, the expected value E[. ] may be determined for a particular time interval (eg, 64 ms) and/or within a particular frequency range (eg, a QMF subband or a number of groups of QMF subbands).

Once the prediction coefficients have been determined, the residual signal D can be determined from the received middle and side signals M, S. The residual signal D can be approximated by the decorrelation form decorr(M) of the received medium signal M. Thus, the noise reduction form S' of the side signal can be determined as S'= a *M+ b *decorr(M), where b is the gain factor for controlling the energy of the decorrelated signal, and is also referred to as the decorrelation parameter b. The decorrelated medium signal decorr (M) can be determined using a decorrelator 78 like the decorrelator 10 of FIG. The de-correlation parameter b can be calculated as b = sqrt(E[D*D]/E[M*M]) in order to control the energy of the same energy as the original residual signal D ( b *decorr(M) )) Replace the residual signal D. As a result, the parameters a, b of the prediction model can be determined in the parameter decision unit 77 from the received medium signal and the received side signal.

As a result, the stereo signals L' and R' are output from the received intermediate signal M and the two parameters a and b at the output of the predictive FM noise reduction system by the noise reduction unit 79. Since parameters a and b are typically estimated and applied in complex-valued QMF domain representations (eg, 64-band), processing can be implemented in a time and frequency manner. Generally, perceptually motivated time and frequency tiling are used. For example, 64 QMF bands can be grouped into a total of 15 frequency bands in accordance with a perceptual frequency scale (eg, a Bark scale). The perceptual frequency scale can be formed by grouping adjacent QMFs at higher frequencies to form a wider frequency band, which is generally referred to as a "parameter band." This set of parameters a and b (for each parameter band) is usually calculated at regular time intervals (time frames), for example using a temporal analysis window of approximately 64 ms length to approximate E[. ] Operation. To ensure a smooth transition of parameter values from one time interval (eg, time frame) to the next, temporal interpolation (eg, linear interpolation along the timeline) can be used to produce a and b. Interpolated parameter values. The interpolated parameter values a and b are then multiplied by the corresponding QMF band signals to which they are applied.

As indicated above, the second QMF analysis row 73 is only used for parameter estimation within the parameter decision unit 77. As can be seen from the formula provided above, the second QMF analysis row 73 provides subband information on the received side signal S, which is used to determine the cross correlation on a per parameter band basis. E[S*M]. In other words, the second QMF analysis row 73 is only used to determine the expected value (in comparison to the QMF frequency band) over the number of stages of the parameter band. In other words, the second QMF analysis row 73 is used to determine the prediction parameters on a relatively coarse time and frequency grid. As a result, the selectivity requirements (e.g., the length of the prototype window), the time/frequency resolution, and/or the second QMF analysis row 73 are significantly less accurate than the QMF analysis band 71 used in the signal path. Demand.

As such, the audio processing device 2 has been described in accordance with FIG. 2 which allows the noise reduction side signal S' to be determined at a reduced computational complexity of the PS-based FM noise reduction system. Using the MS-to-LR converter 76, the side signal S' and the (delayed) received intermediate signal M' can be converted into noise to reduce the left and right stereo signals L', R'. Perceptual testing has shown that in addition to reducing computational complexity, when using the predictive FM noise reduction system outlined herein (e.g., in Figure 3), noise can be improved to reduce the perceived quality of the FM signal.

On the other hand, it has been observed that when a predictive method for FM stereo noise reduction is used, the received signal is dominated by a strong and noisy side signal (ie, having a higher frequency than the middle signal) The number of levels can cause a perceptually annoying human factor. Such a situation can occur, for example, when the transmitted stereo signal is relatively quiet (eg, during a short pause between two pieces of music) while the receiver is facing intermediate to poor reception conditions. Such a situation can be characterized by E[S*S]>>E[M*M], that is, the energy of the received side signal S is (apparently) higher than the energy of the received medium signal M. Since the parameters a and b depend on the energy of the medium signal E[M*M] And partly depending on the fact that the energy of the side signal E[S*S], in the case mentioned above, the parameters a and b usually have large absolute values (obviously greater than 1). This means that the signal M is significantly increased in order to determine the noise reduction side signal S', thereby introducing a human factor. Further, the parameters a and b can fluctuate strongly along time and frequency, which are generally audibly perceived as undesired instability.

To alleviate this problem, a post-processing step can be applied to parameters a and b . In other words, the modified set of parameters a' and b' can be determined by a '=f _a ( a , b ) and b '= f _b ( a , b ). A possible post-processing approach is to apply the attenuation or limiting factor c to obtain post-processing parameters a '= a / c and b '= b / c , where c =1 results in unmodified parameters a and b . The value of c >1 causes the noise reduction side signal S' to be multiplied by 1/ c , which is attenuated by the factor c . It should be noted that other formulas for the relationship between a ', b ' and a , b are possible.

Different techniques are used to calculate the limiting factor c from a and b (i.e., c = f( a , b )). Two different methods are: c = max(1,( a ² + b ² )), or (1)

Use the formula (2) to ensure that the energy of the noise reduction side signal S' does not exceed the energy of the medium signal M, and use the formula (1) to apply even stronger attenuation to the S' in the above case ( Compared to formula (2)), where E[S*S]>E[M*M]. It has been found that the technique using equation (2) tends to provide a slightly better audio quality for wide audio signals in the case of good reception conditions while using the formula The technique of (1) tends to be more reliable in the above-mentioned situation of preventing annoying human factors in the case of intermediate and poor reception conditions.

It should be noted that in the normal reception case, the energy E[S*S] of the side signal is smaller than the energy E[M*M] of the medium signal. In this case, the parameters a and b are usually less than one. The "max" operation in equations (1) and (2) ensures that in such cases the limiting factor is c =1, ie no application restrictions.

As illustrated in FIG. 3, the parameter p can be used to smooth the cross-fading between the noise reduction side signal S' and the originally received (delayed) side signal S in the pass or bypass mode. The pass mode can be beneficial for situations where there are good reception conditions in an optimal manner. For this purpose, the quality of the received FM stereo signal should be estimated in a reliable manner in order to determine the use of a combination of S', S or S' and S for generating noise reduction stereo signals L', R'. In more general terms, the noise reduction side signal S' can be reduced by the noise reduction unit 31 and the bypassed side signal S can pass through the bypass gain unit 30. Gain units 30, 31 generate amplified and/or attenuated side signals from their side signals at their inputs. The amplified and/or attenuated side signals are combined in a merging unit (e.g., adding unit) 32, thereby providing a combined side signal for generating noise reducing stereo signals L', R'.

The predictive FM noise reduction system can further include an HQ (High Quality) detection unit 20 configured to determine or estimate the level of audible noise within the received FM stereo signals L, R (or M, S). number. The estimated number of noise levels determined in the HQ detection unit 20 can be used in the noise The side signal S' is reduced and mixed with the original (bypass) side signal S. In order to mix the side signals, the HQ detecting unit 20 can be configured to set the gain values of the noise reduction gain unit 31 and the bypass gain unit 30. Alternatively or in addition, the mixed side signal can be achieved by interpolating (linear or non-linear) side signals. Alternatively, one of the side signals may be selected based on an estimate of the number of levels of audible noise determined within the HQ detection unit 20.

In the following, the method illustrates how the HQ detecting unit 20 can estimate the actual number of levels of noise in the received FM stereo signal, and thereby decide whether to focus more on the noise reduction side signal S' or more. Bypass side signal S.

In order to distinguish between the noise and the actual payload signal, it is assumed that if the side signal S is significantly stronger than the received medium signal M, the received side signal S mainly contains noise. In other words, it is assumed that if the power of the side signal S exceeds the power of the medium signal M by a predetermined threshold, the power of the side signal S is mainly due to noise. Therefore, the signal-to-noise ratio (SNR) of the received stereo signals M and S can be approximated as a mid-to-side ratio (MSR) for low MSR values: For each frequency band k. MSR_THRESHOLD can be set to, for example, -6dB. In other words, if the energy in the frequency band k of the side signal S The ratio exceeds the energy in the frequency band k of the medium signal M by a predetermined threshold (for example, +6 dB) The MSR can be considered to be equal to or approximate to the SNR in the frequency band k , thereby providing an estimate of the reliable noise contained within the received FM stereo signal.

k =1,..., K frequency bands can be derived, for example, from QMF band analysis stages 71, 73, where K = 64 channels of QMF audio data are available for processing. As outlined above, a QMF or hybrid QMF band can be beneficially clustered into a reduced number of frequency bands that correspond to, for example, a perceptibly motivated scale, such as a Barker scale. In this way, the MSR can be determined for a plurality of frequency (parameter) bands, wherein the resolution of the plurality of frequency bands is perceptually motivated. By way of example, a QMF filter bank can contain 64 QMF bands or a hybrid QMF filter bank can contain 71 bands. The resolution of these filter banks is typically too high in the high frequency range. As such, it may be advantageous to group some of the bands in a perceptually motivated manner. As outlined above, the parameters in the predictive FM noise reduction system correspond to such clustered (perceived) frequency bands. By way of example, the parameters a and b of the predictive FM noise reduction system can be determined using a total of 15 to 20 groups of QMF frequencies within a time window corresponding to a single time frame (including, for example, 2048 samples). The same frequency or parameter band used to determine parameters a and b can also be used to determine the MSR value per frequency/parameter band, thereby reducing overall computational complexity.

The power for the medium signal M and the parameter band k for a given point at time n can be calculated as the expected value: A rectangular window located at a point in time or between samples n ₁ and n ₁ + N - 1 is used. It should be noted that other window shapes can be used to determine the desired value. Alternative time/frequency indications (excluding QMF) can also be used, such as Discrete Fourier Transform (DFT) or other transformations. Also in this case, the frequency coefficients can be grouped into a smaller number of (perceived motivation) parameter bands.

When the side signal S is not stronger than the medium signal M (or is not stronger by the factor MSR_THRESHOLD ), the SNR estimation using the MSR is usually not available. In other words, when the side signal S is not stronger than the medium signal M (or is not stronger by MSR_THRESHOLD ), the MSR is usually not a good estimate of the SNR. In this case, the SNR can be determined based on an estimate of one or more previous SNRs. This can be accomplished by applying a smoothing or attenuation function as illustrated in the context of step 104 of FIG.

4 illustrates the power spectrum for the middle signal 60 and the power spectrum for the side signal 61 in the noisy FM radio reception conditions. For the frequency band of the medium signal M with strong control, it is ambiguous whether the side signal S is noise. The side signal S can be, for example, a partial environmental signal or a partial sway signal. As a result, these frequency bands typically do not provide a reliable indication of the power of the noise within the received FM stereo signal L, R (or M, S). However, looking at the side signal S is a frequency band that is significantly stronger than the medium signal M (for example, at least 6 dB or almost 10 dB), which is considered to be very likely to be essentially pure noise in the side signal S caused by radio transmission. Instructions. Such a situation (where Can be seen in Figure 4 at approximately 2 kHz and 5 kHz. In this way, the minimum value of the MSR across the frequency bands k =1, . . . , K can be regarded as a reliable indicator of the SNR of the received FM radio signal (ie, the quality of the entire received FM radio stereo signal).

Audio content, such as music or speech, typically has less payload energy in the high frequency range than in the lower frequency range. Further In other words, the payload energy in the high frequency range can be less continuous than in the low frequency range. As a result, the energy of the received FM signal noise is easier to detect in the high frequency range than in the low frequency range. In view of this, it may be advantageous to limit the analysis of the MSR to the selected sub-range of the entire K-frequency band. In particular, it may be advantageous to limit the analysis of the MSR to sub-ranges above the entire K frequency band, for example to the upper half of the K frequency band. In this way, the method for detecting the quality of the received FM signal can be made more robust.

In view of the above, a high quality factor α _HQ can be defined which depends on the analysis of the MRS across some or all of the frequency bands k =1, . . . , K (eg, across the high frequency band). The high quality factor α _HQ can be used as an indicator of audible noise within the received FM radio stereo signal. A high quality signal without noise can be indicated by α _HQ =1 and a low quality signal with high noise can be indicated by α _HQ =0. The intermediate quality status can be indicated by 0 < α _HQ <1. The high quality factor α _HQ can be derived from the MSR value based on: The MSR critical MSR_LOW and MSR_HIGH are predetermined normalization thresholds and can be selected in the -6dB and -3dB examples, respectively. Due to this type of normalization, it is ensured that the high quality factor exhibits a value between 0 and 1.

In the above formula, q is a value derived from one or more MSR values. As indicated above, q can be derived from the smallest MSR value across a subset of the frequency bands. Further, q can be set to a peak-decay value of the reverse minimum MSR value. Alternatively or in addition, any other smoothing method can be used to smooth the evolution of the quality indicator parameter q over time.

The high quality factor α _HQ can be used to switch or fade or interpolate between the noise reduction side signal S' and the original unprocessed side signal S. This means that the high quality factor α _HQ = p can be used as the gain for the bypass gain unit 30, however the factor (1- α _HQ ) = 1 - p can be used as the gain for the noise reduction gain unit 31.

An embodiment of the HQ detection algorithm 100 can be illustrated by the steps illustrated in Figure 5 below:

̇ In step 101, the mid- and side-signal powers are calculated, that is, for some or all of the frequencies or parameter bands k , such as K _low < k K _high , the energy of the signal And the energy of the side signal . In one example, K _high = K and K _low = K /2 (ie, only the upper half of the frequency band is considered). Medium and side power and It is decided at time point n , for example using the average formula provided above for the expected value.

In step 102, the mid-to-side ratio (MSR) value for some or all of the frequency bands k is determined to be, for example,

̇ In step 103, determining the MSR value for a certain frequency range , where the frequency range is, for example, K _low < k K _high .

步骤 In step 104, the minimum MSR value is smoothed by time γ _peak ( n )=min( κγ _peak ( n −1), γ _min ), for example, by determining the MSR peak as the following expression, with attenuation The factor κ = exp(-1/( F _s τ )) has a time constant of, for example, τ = 2 seconds, and has F _s as a frame rate, that is, a ratio of how high the frequency of step 104 is. This establishes a reverse peak attenuation that smoothes the minimum MSR value over time.

˙ In step 105, the high quality factor at time n by the use of α _HQ based MSR peak γ _peak (n) at time n decision, i.e., by using the time point n smooth The minimum MSR value γ _peak ( n ), with q = γ _peak ( n ) As indicated above, the MSR threshold can be set to, for example, MSR_LOW = -6 dB and MSR_HIGH = -3 dB.

In step 107, the high quality factor α _HQ at time point n can be applied to the side signal mixing process illustrated in FIG.

The HQ detection algorithm 100 mentioned above can be repeated for subsequent time points (as illustrated by the arrow from step 107 back to step 101).

The method or system for determining the high quality of the received FM radio stereo signal may further dictate that the high quality factor α _HQ depends on one or more additional indicators (in addition to one or more MSR values). In particular, the high quality factor can be made dependent on the Spectral Flatness Measurement (FSM) of the received FM radio stereo signal. The so-called SFM_impact_factor , which is normalized between 0 and 1, can be determined as outlined in WO PCT/EP2011/ 064077 . SFM_impact_factor =0 may correspond to a low SFM value indicating the power spectrum of the side signal S for which the spectral power is concentrated in a relatively small number of signal bands. That is, the SFM impact factor "0" indicates low-level noise. On the other hand, the SFM impact factor "1" corresponds to a high SFM value indicating that the spectrum has a similar amount of power in all spectral bands. As a result, the SFM impact factor "1" indicates the noise of the advanced number.

The modified high quality factor α ' _HQ can be determined according to the following formula: This emphasizes the high quality factor α ' _HQ =0 (indicating low quality, ie high noise) if SFM_impact_factor =1 (indicating high-level noise in the received FM radio stereo signal) and vice versa . It should be noted that the above formula for combining the effects of MSR-based high quality factors α _HQ and SFM is only one possible combination of the (modified) high quality factor α ' _HQ combined with two noise indicators. method. The SFM_impact_factor may be a noise that facilitates detecting a situation in which both the neutral and side signals have a fairly flat spectrum and are close in energy. In such cases, the minimum MSR value γ _min is typically close to 0 dB despite a significant amount of audible noise within the received FM stereo signal. In the above-described PS processing/bypass mixing processing, the modified high quality factor α ' _HQ can replace the high quality factor α _HQ .

In the following, another option for enhancing the method and system for HQ detection is illustrated. The modified high quality factor can be determined by the high quality factor α _HQ affecting the total number of side sums S _sum into a soft noise gate, that is, the total number of side signals (ie, energy or power). It can be determined as the energy of the side signal (across all frequency bands). As a result, the modified high quality factor α ' _HQ can be determined according to the following formula: The critical S_THRES_LOW and S_THRES_HIGH can be used to normalize the gate factor g _gate to a value between 0 and 1. An FM signal with a sequence S _sum < S_THRES_LOW is considered low quality, whereas an FM signal with a sequence S _sum > S_THRES_HIGH can be of high quality.

Another selection system for providing enhanced detection algorithm of the HQ of the high quality factor [alpha] _HQ detector concealed by, for example, as described in WO PCT / EP2011 / 064084 in (concealment detector) output of the impact. The modified high quality factor α ' _HQ can be determined by considering whether the concealment within the predictive FM radio reduction system is active in order to conceal the mono drop condition of the unwanted FM receiver. The modified high quality factor α ' _HQ can be determined according to α ' _HQ = (1 - δ _conceal ) α _HQ , where δ _conceal =1 if concealment is active, and δ _conceal =0 otherwise. This means that the received FM radio signal is of course considered low quality ( α ' _HQ =0) if it is concealed as active in the predictive FM radio noise reduction system, otherwise the quality of the received FM radio signal is based on a high quality factor. The calculated value of α _HQ is estimated. In order to avoid (audible) discontinuities when restoring from the concealed state (ie δ _conceal =1), ie to ensure that the modified high quality factor α ' _{HQ is} smoothed from 0 to a non-zero value, whenever When δ _conceal =1, the minimum MSR value γ _min can be forced to γ _min = MSR_LOW , so that the smooth transition is ensured by the smoothing method of step 104 of FIG. 5 . Since the high quality factor depends on the concealed state δ _conceal , it is possible to establish a fast switching of the mode using the predicted FM radio noise reduction (that is, the FM radio noise reduction processing for sudden occurrence of bad reception conditions is fast) Transfer), and slow mix back to bypass mode (when the receiving conditions have improved).

In the following, another option for enhancing the HQ detection method will be described. The MSR value γ _k can be adjusted for large sway signals based on: The parameter λ indicates the degree to which the received FM radio stereo signal is swayed. The parameter λ can be determined from the ratio of the received left signal L and the received right signal R energy, for example, based on: The energy or power with P ^L = E { L ² } as the received left signal and P ^R = E { R ² } as the energy or power of the received right signal. As a result, the MSR value γ _k is increased for a sway signal having a significant energy difference between the left signal L and the right signal R. This large difference between the L and R signals causes the side signal S to have a relatively high energy, although the side signal S does not contain noise. By increasing the MSR value γ _k , the minimum MSR value γ _{min is} increased, thereby increasing the high quality factor α _HQ . As a result, the use of the parameter λ helps to avoid false detection of low quality signals from the strong side signal S due to the wide (music) stereo mixing and stereo widening post processing.

It should be noted that the selection of the high quality factor α _HQ proposed above for determining the modification can be used singly or in any combination with each other.

Further, it should be noted that the high quality factor α _HQ can be used to adjust parameters a and b in a predictive FM stereo radio noise reduction system. In particular, the limit factor c can be affected by the quality indicator α _HQ . This can be done, for example, according to the following formula: Where ε is the selected adjustment value (small number) to prevent a and b from being infinite (or unreasonably large) when the quality indicator α _HQ =1, that is, when the received FM signal contains a low degree of noise .

The purpose of the limit function c = f( a , b , α _HQ ) depending on the quality indicator α _HQ is to limit a and b for low quality FM signals ( α _HQ close to 0), while not (or only slight) ) Limit a and b for high quality FM signals ( α _{HQ is} close to 1). It should be noted that the above-mentioned function for modifying the limiting factor depending on the quality indicator α _HQ approximates the first function (1) for c with α _HQ =0, the second function (2) for α _HQ = 0.5, and α _HQ =1 applies the “unrestricted” parameters a and b . Further, it should be noted that the above-mentioned formula is only an example of a modified limit function for constructing consideration of the quality of the received FM signal.

The selection and combining of the noise reduction side signal S' and the bypass side signal S illustrated in FIG. 3 can be performed in a frequency selective manner. Possible builds will include the following modifications to the block diagram of Figure 3. The block diagram in Figure 3 can be modified such that gain units 30, 31 and merging unit 32 are implemented in the QMF domain prior to side signal synthesis filter bank "QMF ^-1 " 72. Further, the input to the bypass gain unit 30 may be the output of the "QMF _s " analysis filter bank 73. This would mean that the filter bank 73 is in the signal path in the case of passing, and therefore has the same accuracy requirements as the "QMF" analysis filter bank 71. The QMF synthesis filter bank 72 can be used to convert the combined side signals (downstream of the merging unit 32) into a time domain.

In an alternative embodiment, the frequency is selectively limited to two frequency bands, namely a high frequency band and a low frequency band. In particular, the low frequency band can be fixed to the bypass path, that is, the reconstructed side signal can correspond to the side signal S for low frequency range reception, but in the high frequency range, the side signal S can be reduced by using noise. '(Or according to the mixed side signal of the quality indicator p).

WO PCT/EP2011/064077 describes techniques for reducing or removing the amount of uncorrelated components that are undesirable in noise reduction stereo signals by using a spectral flatness measure. These techniques can also be applied to the predictive FM radio noise reduction system described in this document. In particular, the spectral flatness measurement can be applied by modifying the parameter b as follows: b _new=(1-SMF_impact_factor)* b . This means that SFM_impact_factor=1 forces b _new=0. For SFM_impact_factor=0, b will remain unchanged. In this way, in the side signal (with a noisy side signal) with high spectral flatness with SFM_impact_factor=1, no de-correlation phase is added to the noise reduction side signal S', so that the noise reduction side signal S' corresponds to To the scaled version of the received signal, that is, a *M.

In the following, an overview is given for the example of determining SFM_impact_factor . In the normally received FM radio stereo signal, the power spectrum of the medium signal M is relatively steep in the lower frequency range and has the energy of the advanced number. On the other hand, the signal S typically has an overall low degree of energy and a relatively flat power spectrum.

Since the power spectrum of the side signal noise is quite flat and has a characteristic slope, the slope compensated SFM can be used to estimate the number of noise levels within the received FM signal. Different types of SFM values can be used. That is, the SFM value can be calculated in various ways. In particular, instantaneous SFM values and smoothed versions of SFM can be used. The instantaneous SFM value usually corresponds to the SFM of the signal frame of the side signal, however the smoothed version of the instantaneous SFM value also depends on the SFM of the previous signal frame of the side signal.

The method for determining the impact factor from the side signal includes the step of determining the power spectrum of the side signal. Typically, this is a certain number of samples using the side signal (for example, the sampling of the signal frame). The power spectrum can be determined as the energy value of the side signal , for a plurality of frequency bands k , such as k =1,..., K . The decision period of the power spectrum can be aligned with the period used to determine parameters a and b . In this way, the power spectrum of the side signal can be determined according to the effective period of the corresponding parameters a and b .

In the subsequent steps, the characteristic slope of the power spectrum of the side signal noise can be compensated. The characteristic slope (in the design/tuning phase) can be experimentally determined, for example, by determining the average power spectrum of the side signals of a set of mono signals. Alternatively or in addition, adaptability determines the slope of the characteristic from the current side signal, for example using linear regression on the power spectrum of the current side signal. Compensation of characteristic slope can be reversed noise slope The inverse noise slope filter is implemented. As a result, a slope-compensated, possibly flat power spectrum should be obtained which does not show the characteristic slope of the power spectrum of the side signal of the monophonic speech audio signal.

The SFM value can be determined using the (slope compensated) power spectrum. SFM can be calculated according to the following formula among them It means the power of the side signal in the frequency or parameter with k . Frequency partitioning used in predictive FM noise reduction systems typically includes 15 to 20 parameter bands. The SFM can be illustrated as a radio between the geometric mean of the power spectrum and the arithmetic mean of the power spectrum.

Alternatively, the SFM can be computed over a subset of the spectrum, including only frequency bands ranging from K _low to K _high . This approach, for example one or a few frequency bands, can be excluded in order to remove unwanted DC (eg low frequency) offsets. When adjusting the band boundary, the formula proposed above for calculating the SFM should be modified accordingly.

In order to limit the computational complexity, the SFM formula can be approximated by a numerical expansion based on Taylor expansion, a look-up table, or a similar technique generally known to those skilled in the art of software construction. (numerical approximation) replaced. Further, there are other methods of measuring spectral flatness, such as, for example, the difference between the minimum and maximum of a standard deviation or a frequency power bin. In this document, the term "SFM" means any of these measurements. By.

The impact factor can be determined using the SFM value for the time frame of a particular time period or side signal. For this purpose, the SFM is mapped to a scale of, for example, 0 to 1. The mapping and decision of the SFM impact factor can be determined according to the following formula Two of the threshold values α _{low_thresh} and α _{high_thresh} are selected according to the average range of SFM values generally ranging from 0.2 to 0.8. The main purpose of the normalization level is to ensure that the SFM impact factor is between the "0" and "1". As a result, normalization ensures that the "normal" non-flat spectrum (SFM < α _{low_thresh} ) is not detected as noise and the measurement is saturated for high values (SFM > α _{high_thresh} ). In other words, normalization provides a clearer difference between the high noise situation (SFM> α _{high_thresh} ) and the low noise condition (SFM< α _{low_thresh} ).

WO PCT/EP2011/064084 describes a technique for concealing the short interval of the mono reception of the FM receiver 1 by means of a reliable mono-detector in combination with the previously estimated stereo parameters. Noise is generated during the class-like time interval to reduce the FM stereo signal. The technique outlined in WO PCT/EP2011/064084 can also be applied to the predictive FM radio noise reduction system described in this document.

As indicated above, the FM receiver 1 can toggle between stereo and mono due to time-variant poor reception conditions (eg, "fading"). To maintain stereo sound image during mono/stereo switching, error concealment techniques can be used to conceal short mono downsizing. The hidden method in predictive FM radio noise is to use the prediction and decorrelation parameters a and b , which are based on previously estimated parameters, in case the audio output of the FM receiver 1 has been lowered to mono and cannot be calculated new. The case of parameters a and b . Thus, when the FM stereo receiver 1 switches to the mono audio output, the predictive FM radio noise reduction system of Figure 3 continues to use the previously estimated parameters a and b (individual for each frequency band). If the fallback period in the stereo output is short enough that the stereo image of the FM radio signal remains similar during the fallback period, the fallback in the audio output of device 2 is inaudible or only barely audible. . Another approach may be to interpolate and/or extrapolate parameters a and b from previously estimated parameters. If the FM reception is not fast enough to return to stereo, the parameters a and b can be slowly attenuated to approach 0 after a few times, which means that the mono signal (i.e., the medium signal) is output.

Alternatively or in addition, the predictive FM stereo noise reduction system may parameterize a and/or b to generate a "false stereo" signal using a default value to prevent the reception condition from being so bad that only mono signals are received. situation. The default value can depend on the voice/music classification of the medium signal. In other words, the predictive FM stereo noise reduction system can include a classifier for classifying the type of received FM radio signal based on the received medium signal. By way of example, the classifier can be configured to classify the received FM radio signal into a voice signal or a music signal (eg, based on the frequency analysis of the received medium signal). The predictive FM stereo noise reduction system can then select an appropriate value for a and/or b based on the type of decision of the received FM radio signal. As a result, the mono drop of the received FM radio signal can be concealed using (type dependent) default parameter values.

The use of concealment in predictive FM radio noise reduction systems requires reliable detection of mono-destination in order to trigger concealment, ie to set the concealed state δ _conceal from 0 to 1. A possible mono/stereo detector can be based on a mono segment of the detected signal, which satisfies the condition of left signal = right signal (or left signal - right signal = 0). However, such mono/stereo detectors can cause unstable behavior for concealed processing because the left and right signal energies and side signal energies can vary widely (even under sound reception conditions). fact.

In order to avoid such hidden and unstable behavior, the mono/stereo detection and concealment mechanism can be built as a state machine. An exemplary state machine is illustrated in Figure 6. The state machine of Figure 6 utilizes two reference levels of the absolute energy of the side signal S, namely E _S (or P ^S as defined above). The side signal S used to calculate E _s may be high pass filtered with a cutoff frequency of typically 250 Hz. These reference series are the upper reference level ref_high and the lower reference level ref_low. The signal is considered stereo above the upper reference level (ref_high) and mono in the lower reference level (ref_low).

The side signal energy E _S is calculated as the control parameter of the state machine. E _S can be calculated over a period of time window, which can for example correspond to a time period of validity of parameters a and b . In other words, the frequency at which the side signal energy is determined can be aligned with the frequency at which parameters a and b are determined. In this document, the time period for determining the side signal energy E _S (and possibly the parameters a and b ) is referred to as the signal time frame. The state machine of Figure 6 contains five conditions (which are verified each time a new time frame energy E _{S is} calculated):

- Condition A indicates that the side signal energy E _S exceeds the upper reference level ref_high. The upper reference series can be referred to as a higher criticality.

- Condition B indicates that the side signal energy E _{S is} lower than or equal to the upper reference level ref_high and higher than or equal to the lower reference level ref_low. The lower reference series can be referred to as a lower critical.

- Condition B1 corresponds to condition B, but additional time conditions are added. The time condition specifies that condition B satisfies a critical number less than the time frame or less than the critical time. This criticality can be referred to as the time frame critical.

- Condition B2 corresponds to condition B, with an additional time condition that specifies that condition B satisfies the critical number of boxes greater than or equal to or greater than or equal to the critical time.

- Condition C indicates that the side signal energy E _{S is} lower than the lower reference level ref_low.

Further, the example state machine of Figure 6 utilizes five states. The different states are subject to the conditions mentioned above and are reached by the state diagrams illustrated in Figure 6. The following actions are typically performed in different states within the predictive FM radio stereo noise reduction system:

- In state 1, a normal stereo operation is performed, for example based on parameters a and b determined from the current audio signal. The concealed state δ _conceal remains at zero.

- In State 2, normal stereo operation is performed based on parameters a and b determined on the current audio signal. This state is only transitional, since the frame is critical when the number of time frames is greater than or equal to, or the time is greater than or equal to the time threshold (ie, condition B2) or before the elapse of time or the number of frames at this time. If the fact that the state B is satisfied for a while, the condition A or C is satisfied. The concealed state δ _conceal remains at zero.

- In state 3, stereo operation is performed based on parameters a and b determined on the current audio signal. It can be seen that state 3 can be reached on the path from state 1 to state 3 via state 2. In view of the fact that condition B2 requires a minimum number of time frames or a minimum number of times for transfer, the path "State 1, State 2, State 3" represents from normal stereo operation (eg, music) to normal mono operation (eg, Slow (ie smooth) transfer of speech). The concealed state δ _conceal is set at or maintained at zero.

- In state 4, the mono-reduction concealment begins with the previously determined parameters a and b , such as the most recent parameters a and b that were determined in state 1. It can be seen that if the condition C is satisfied, that is, if the side signal energy E _S falls steeply from ref_high or more to less than ref_low, the state 4 can be directly reached from the state 1. Alternatively, however, state 4 can be reached from state 1 via state 2 only if condition B is satisfied only for a small number of time frames or only for a short cycle time. As a result, the paths "state 1, state 4" and "state 1, state 2, state 4" represent a fast, ie abrupt transition from, a normal stereo operation (eg, music) to a forced mono operation. If the series or intensity of the 19 kHz pilot tone falls below a predetermined number of levels in the stereo multiplex signal, the forced mono operation is usually received by the FM receiver, for example, by a sudden cut-off side signal. It is impossible to make a stereo demodulation signal to make a reliable demodulation side signal. The concealed state δ _conceal is set to 1 to indicate the use of concealment within the predictive FM radio noise reduction system.

- In state 5, for example based on the parameters a and b that have been established in state 4, the mono drop concealment is continued. In the illustrated embodiment, if state C is satisfied, state 5 can only be reached from state 4, that is, state 5 represents a stable mono drop concealed state in which previously determined parameters a and b are used in order to The received medium signal generates a noise reduction side signal. The parameters a and b can decay to zero with a time constant of a few seconds, resulting in a slow transition from stereo to mono output. The hidden state δ _conceal is usually kept at 1.

As indicated, the state diagram illustrated ensures that only the audio signal received by the FM receiver goes from stereo to mono in a small time window/time frame, ie if the transition from stereo to mono is sudden. , triggers concealment. On the other hand, in the case where there is noise in the side signal having the energy level E _S lower than the stereo level (ref_high) but higher than the mono level (ref_low), that is, in the side signal In the case where there is still sufficient information to generate the appropriate a and b , it is prevented from triggering concealment. At the same time, even when the signal changes from stereo to mono, such as when the signal transitions from music to speech, it will not trigger covert detection, thereby ensuring that the original mono signal is not presented due to erroneous application concealment. Become an artificial stereo signal. Based on the smooth transition from the ref_high to the lower signal energy E _S below the ref_low, the true transition from stereo to mono can be detected.

In this document, methods and systems for improving the perceived performance of an FM radio receiver have been described. In particular, methods and systems for determining noise reduction FM stereo signals using predictive techniques have been described. By using a predictive FM radio noise reduction system, the computational complexity for noise reduction can be reduced compared to a PS-based FM radio noise reduction system. Further, various methods for improving the performance of predictive FM radio noise reduction systems have been described. In particular, the use of a quality indicator has been described to mix between the noise reduction side signal and the original side signal. Further, a method for applying the parameters of the predictive FM radio noise reduction system to the spectral characteristics of the received side signals has been described, thereby reliably distinguishing between noisy and good reception conditions. In addition, the concealment method has been described in order to apply the predictive FM radio noise reduction system to the mono drop condition.

The methods and systems described in this document can be implemented as software, firmware, and/or hardware. Some components may be implemented, for example, as software running on a digital signal processor or microprocessor. Other components may be implemented, for example, as hardware and or application specific integrated circuits. The signals encountered in the described methods and systems can be stored on the media, such as random access memory or optical storage media. They can be transmitted via the Internet, such as radio networks, satellite networks, wireless Network or wired network (such as the Internet). A typical device utilizing the methods and systems described in this document is a portable electronic device or other consumer device for storing and/or presenting audio signals.

2‧‧‧ Equipment

20‧‧‧Detection unit

30‧‧‧Bypass gain unit

31‧‧‧ Noise reduction gain unit

32‧‧‧Merge unit

71‧‧‧Filter row

72‧‧‧ Filter row

73‧‧‧Filter row

74‧‧‧Delay

75‧‧‧LR-to-MS converter

76‧‧‧MS-to-LR converter

77‧‧‧Parameter decision unit

78‧‧‧Resolver

79‧‧‧ Noise Reduction Unit

80‧‧‧ thin line

81‧‧‧ thick line

82‧‧‧ dotted line

Claims (34)

A device (2) configured to reduce noise of a received multi-channel FM radio signal; wherein the received multi-channel FM radio signal can be represented as a received medium signal and a received side signal; wherein the receiving side The signal indicates the difference between the left signal and the right signal of the received multi-channel FM radio signal; the device (2) includes a parameter determining unit (77) configured to determine the received medium signal and the reception One or more parameters related to the correlation and/or decorrelation between the side signals; wherein the parameter decision unit (77) is configured to determine a de-correlation between the received intermediate signal and the received side signal Decoupling parameter b ; and - a noise reduction unit (79) configured to generate a noise reduction side signal from the received medium signal using the one or more parameters; wherein the noise reduction unit (79) is configured The use of the decorrelation parameter b also generates the noise reduction side signal from the de-correlated form of the received medium signal; wherein the received side signal is not in the signal path for generating the noise reduction side signal. The apparatus (2) of claim 1, wherein the parameter determining unit (77) is configured to determine a prediction parameter a indicating a cross correlation between the received intermediate signal and the received side signal; - the noise reduction unit (79) is configured to generate the noise reduction side signal from the received intermediate signal using the prediction parameter a . The device (2) of claim 2, wherein the parameter determining unit (77) is configured to determine the prediction parameter based on an expected value of a product of the received medium signal and a corresponding sample of the received side signal. a . For example, the device (2) of claim 3, wherein the parameter determining unit (77) is configured to determine the prediction parameter a as a = E[S*M]/E[M*M], where E [. ] means the desired operand, S is the received side signal and M is the received medium signal. The device (2) of claim 2, wherein the parameter determining unit (77) is configured to determine based on the received side signal and the energy of the difference signal of the signal determined from the medium signal using the prediction parameter a. The decorrelation parameter b . For example, the device (2) of claim 5, wherein the parameter determining unit (77) is configured to determine the de-correlation parameter b as b = sqrt(E[D*D]/E[M*M] ) has D = S - a * M as the difference signal. For example, the device (2) of claim 4, wherein the noise reduction unit (79) is configured to generate the decoupling form of the received intermediate signal by filtering the received intermediate signal by using an all-pass filter. . For example, the device (2) of claim 4, wherein - the parameter determining unit (77) is configured to determine the characteristic of the impact factor of the spectral flatness of the received side signal; and - the decorrelation parameter b Depends on the impact factor. The device (2) of claim 8 wherein the correlation parameter b decreases when the impact factor indicates an increase in the spectral flatness of the received side signal. For example, the device (2) of claim 1 of the patent scope, wherein - the parameter decision unit (77) is configured to determine the one or more parameters in a time variable manner; and - the noise reduction unit (79) is a group The parameter is used to generate the noise reduction side signal using parameters of the one or more time variables. The apparatus (2) of claim 10, wherein - the parameter determining unit (77) is configured to determine, for the first one of the one or more parameters, a first parameter for a corresponding sequence of time intervals a sequence; and - using a sample of the received intermediate signal and/or the received side signal during a particular time interval, determining a particular first parameter of the sequence of the first parameter for the particular time interval of the sequence of time intervals . The device (2) of claim 11, wherein the parameter determining unit (77) is configured to determine the first parameter of the interpolation by interpolating the adjacent first parameter from the sequence of the first parameter. The sequence. The device (2), as claimed in claim 10, is configured to detect that the received multi-channel FM radio signal is a forced mono signal, the detection being by - in the sequence of time intervals Determining the energy of the received side signal during the first time interval; wherein the energy is above the high threshold; - determining a number of transition periods following a continuous time interval during which the energy of the side signal is higher than the high Critical landing below low critical a value; and - determining that if the number of consecutive time intervals of the transition period is below the interval threshold, the received multi-channel FM radio signal following the first time interval is a forced mono signal. The device of claim 13, wherein the parameter determining unit is determined if the received multi-channel FM radio signal is a forced mono signal in a time interval following the first time interval. The system is configured to determine the one or more parameters for the time interval following the first time interval from the one or more parameters for the first time interval. For example, the device (2) of claim 1 of the patent scope, wherein - the parameter decision unit (77) is configured to determine the one or more parameters in a frequency variable manner; and - the noise reduction unit (79) is a group The state is generated by using the one or more frequency variable parameters to generate the noise reduction side signal. The device (2) of claim 15 further comprising a medium conversion unit (71) configured to generate a plurality of neutron band signals covering the plurality of frequency ranges from the received medium signal; and - side a transform unit (73) configured to generate a plurality of side subband signals covering the corresponding plurality of frequency ranges from the received side signal; and wherein the parameter decision unit (77) is configured to configure the one or more The second parameter determines a plurality of second subband parameters from the corresponding plurality of neutron band signals and the corresponding plurality of side subband signals. The apparatus (2) of claim 16 wherein the side transform unit (73) satisfies a lower requirement than the medium transform unit (71) in at least one of: - frequency selectivity; Frequency resolution; - time resolution; and - numerical precision. For example, the device (2) of claim 16 wherein the noise reduction unit (79) is configured to generate a complex number from the corresponding plurality of neutron band signals and the corresponding plurality of second subband parameters The noise reduction side subband signal; and - the noise reduction unit (79) includes an inverse transform unit (72) configured to generate the noise reduction side signal from the plurality of noise reduction side subband signals. The device (2) of claim 16 wherein the medium transform unit (71) and/or the side transform unit (73) is a QMF filter bank. The apparatus (2) of claim 1, wherein the parameter decision unit (77) is configured to limit the one or more parameters by applying a limit factor c to the one or more parameters. The apparatus (2) of claim 20, wherein for c > 1, the limiting factor c is proportional to the sum of the one or more squared parameters. The apparatus (2) of claim 20, wherein, for c > 1, the limiting factor c is proportional to the square root of the sum of the one or more squared parameters. For example, the device (2) of claim 20, wherein the application of the limit factor c does not increase the one or more parameters. For example, the device (2) of claim 1 includes a delay unit (74) configured to delay the sampling of the received intermediate signal to correspond to the calculation of the corresponding sample required to generate the noise reduction side signal. The amount of time. The device (2) of claim 1 includes a combining unit (30, 31, 32) configured to use a quality indicator indicating the quality of the received multi-channel FM radio signal from the noise reduction side The signal and the received side signal determine the modified noise reduction side signal. For example, the device (2) of claim 25, wherein the combining unit (30, 31, 32) includes a noise reduction gain unit (31) configured to reduce noise by using noise to reduce the side signal Weighting; - a bypass gain unit (30) configured to weight the received side signal using a bypass gain; and - a merging unit (32) configured to combine the weighted noise reduction side signal with the weighted The received side signal; wherein the noise reduction gain and the bypass gain are dependent on the quality indicator. For example, the device (2) of claim 25, wherein the combining unit is configured to determine the modified noise reduction side signal in a frequency selective manner. The device (2) of claim 25, comprising a quality determining unit (20) configured to determine the quality indicator, the decision determining (101) the power of the received medium signal by means of - Referring to the medium power, and determining the power of the received side signal, which is referred to as side power; determining (102) the ratio of the medium power to the side power, thereby producing a mid-side ratio; and - based on at least The mid-to-side ratio determines (105) the quality indicator of the received FM radio signal. The device (2) of claim 1 includes an MS-to-LR converter (76) configured to determine a noise from the received middle signal and the noise reduction side signal to reduce left signal and miscellaneous The message lowers the right signal. The device (2) of claim 29, wherein the MS-to-LR converter (76) is configured to determine - the noise from the sum of the received medium signal and the noise reduction side signal Decreasing the left signal; and - reducing the right signal from the noise of the received intermediate signal and the noise reduction side signal. A method for reducing noise of a received multi-channel FM radio signal; wherein the received multi-channel FM radio signal can be presented as a received intermediate signal and a received side signal; the method includes - determining a representation of the reception One or more parameters related to the correlation and/or de-correlation between the received signal and the received side signal; wherein the one or more parameters include a de-correlation between the received intermediate signal and the received side signal Deciding the parameter b ; and - using the one or more parameters to generate a noise reduction side signal from the received medium signal, wherein generating the noise reduction side signal comprises using the correlation parameter b also from the received medium signal The decorrelation form generates the noise reduction side signal; wherein the received side signal is not in the signal path for generating the noise reduction side signal. A software program adapted to be executed on a processor and adapted to perform the method steps of claim 31 when implemented on a computing device. A storage medium comprising a software program adapted to be executed on a processor and adapted to perform the method steps of claim 31 when implemented on a computing device. A computer program product comprising executable instructions for performing the method steps of claim 31 when executed on a computer.