EP1845519B1

EP1845519B1 - Encoding and decoding of multi-channel audio signals based on a main and side signal representation

Info

Publication number: EP1845519B1
Application number: EP07109801A
Authority: EP
Inventors: Ingemar Johansson; Anisse Taleb; Stefan Bruhn; Daniel ENSTRÖM
Original assignee: Telefonaktiebolaget LM Ericsson AB
Current assignee: Telefonaktiebolaget LM Ericsson AB
Priority date: 2003-12-19
Filing date: 2004-12-15
Publication date: 2009-09-16
Anticipated expiration: 2024-12-15
Also published as: WO2005059899A1; CA2690885C; JP2008026914A; BRPI0410856B1; CN101118747A; PL1623411T3; ATE371924T1; SE527670C2; JP2007529021A; HK1091585A1; DE602004008613T2; SE0400417L; ATE443317T1; BRPI0419281B1; EP1623411B1; CN101118747B; CA2690885A1; MXPA05012230A; JP4589366B2; DE602004008613D1

Abstract

A method of encoding multi-channel audio signals comprises generating of a first output signal (x' mono ), being encoding (38) parameters representing a main signal (x mono ). The main signal (x mono ) is a first linear combination (34) of signals (16A,16B) of at least a first and a second channel. The method further comprises generating (30) of a second output signal (p side ), being encoding parameters representing a side signal (x side ). The side signal (x side ) is a second linear combination (36) of signals (16A,16B) of at least the first and the second channel within an encoding frame. The method is characterised in that the generating of the second output signal further comprises scaling of the side signal (x side ) to an energy contour of the main signal (x mono ). A method of decoding is also presented as well as an encoder, a decoder and audio system, all according to the same basic idea.

Description

TECHNICAL FIELD

The present invention relates in general to encoding of audio signals, and in particular to encoding of multi-channel audio signals.

BACKGROUND

There is a high market need to transmit and store audio signals at low bit rate while maintaining high audio quality. Particularly, in cases where transmission resources or storage is limited low bit rate operation is an essential cost factor. This is typically the case, e.g. in streaming and messaging applications in mobile communication systems such as GSM, UMTS, or CDMA.
Today, there are no standardised codecs available providing high stereophonic audio quality at bit rates that are economically interesting for use in mobile communication systems. What is possible with available codecs is monophonic transmission of the audio signals. To some extent also stereophonic transmission is available. However, bit rate limitations usually require limiting the stereo representation quite drastically.
The simplest way of stereophonic or multi-channel coding of audio signals is to encode the signals of the different channels separately as individual and independent signals. Another basic way used in stereo FM radio transmission and which ensures compatibility with legacy mono radio receivers is to transmit a sum and a difference signal of the two involved channels.
State-of-the-art audio codecs, such as MPEG-1/2 Layer III and MPEG-2/4 AAC make use of so-called joint stereo coding. According to this technique, the signals of the different channels are processed jointly, rather than separately and individually. The two most commonly used joint stereo coding techniques are known as "Mid/Side" (M/S) stereo coding and intensity stereo coding, which usually are applied on sub-bands of the stereo or multi-channel signals to be encoded.
M/S stereo coding is similar to the described procedure in stereo FM radio, in a sense that it encodes and transmits the sum and difference signals of the channel sub-bands and thereby exploits redundancy between the channel sub-bands. The structure and operation of an encoder based on M/S stereo coding is described, e.g. in US patent 5,285,498 by J.D. Johnston .
Intensity stereo on the other hand is able to make use of stereo irrelevancy. It transmits the joint intensity of the channels (of the different sub-bands) along with some location information indicating how the intensity is distributed among the channels. Intensity stereo does only provide spectral magnitude information of the channels. Phase information is not conveyed. For this reason and since the temporal inter-channel information (more specifically the inter-channel time difference) is of major psycho-acoustical relevancy particularly at lower frequencies, intensity stereo can only be used at high frequencies above e.g. 2 kHz. An intensity stereo coding method is described, e.g. in the European patent 0497413 by R. Veldhuis et al .
A recently developed stereo coding method is described, e.g. in a conference paper with the title "Binaural cue coding applied to stereo and multi-channel audio compression", 112th AES convention, May 2002, Munich, Germany by C. Faller et al. This method is a parametric multi-channel audio coding method. The basic principle is that at the encoding side, the input signals from N channels c₁, c₂, ... c_N are combined to one mono signal m. The mono signal is audio encoded using any conventional monophonic audio codec. In parallel, parameters are derived from the channel signals, which describe the multi-channel image. The parameters are encoded and transmitted to the decoder, along with the audio bit stream. The decoder first decodes the mono signal m' and then regenerates the channel signals c₁', c₂',..., c_N', based on the parametric description of the multi-channel image.
The principle of the Binaural Cue Coding (BCC) method is that it transmits the encoded mono signal and so-called BCC parameters. The BCC parameters comprise coded inter-channel level differences and inter-channel time differences for sub-bands of the original multi-channel input signal. The decoder regenerates the different channel signals by applying sub-band-wise level and phase adjustments of the mono signal based on the BCC parameters. The advantage over e.g. M/S or intensity stereo is that stereo information comprising temporal inter-channel information is transmitted at much lower bit rates. However, this technique requires computational demanding time-frequency transforms on each of the channels, both at the encoder and the decoder.
Moreover, BCC does not handle the fact that a lot of the stereo information, especially at low frequencies, is diffuse, i.e. it does not come from any specific direction. Diffuse sound fields exist in both channels of a stereo recording but they are to a great extent out of phase with respect to each other. If an algorithm such as BCC is subject to recordings with a great amount of diffuse sound fields the reproduced stereo image will become confused, jumping from left to right as the BCC algorithm can only pan the signal in specific frequency bands to the left or right.
A possible means to encode the stereo signal and ensure good reproduction of diffuse sound fields is to use an encoding scheme very similar to the technique used in FM stereo radio broadcast, namely to encode the mono (Left+Right) and the difference (Left-Right) signals separately.
A technique, described in US patent 5,434,948 by C.E. Holt et al. uses a similar technique as in BCC for encoding the mono signal and side information. In this case, side information consists of predictor filters and optionally a residual signal. The predictor filters, estimated by a least-mean-square algorithm, when applied to the mono signal allow the prediction of the multi-channel audio signals. With this technique one is able to reach very low bit rate encoding of multi-channel audio sources, however, at the expense of a quality drop, discussed further below.
Finally, for completeness, a technique is to be mentioned that is used in 3D audio. This technique synthesises the right and left channel signals by filtering sound source signals with so-called head-related filters. However, this technique requires the different sound source signals to be separated and can thus not generally be applied for stereo or multi-channel coding.

SUMMARY

A problem with existing encoding schemes based on encoding of frames of signals, in particular a main signal and one or more side signals, is the existence of the pre-echo effect. In Fig. 7a-b, diagrams are illustrating such an artefact. Assume a signal component having the time development as shown by curve 100. In the beginning, starting from t0, the signal component is not present in the audio sample. At a time t between t1 and t2, the signal component suddenly appears. When the signal component is encoded, using a frame length of t2-t1, the occurrence of the signal component will be "smeared out" over the entire frame, as indicated in curve 101. If a decoding takes place of the curve 101, the signal component appears a time Δt before the intended appearance of the signal component, and a "pre-echo" is perceived.
An object of the present invention is therefore to provide an encoding method and device improving the perception quality of multi-channel audio signals, in particular to avoid artefacts such as pre-echoing. A further object of the present invention is to provide an encoding method and device requiring less processing power and having more constant transmission bit rate requirements.
The above objects are achieved by methods and devices according to the enclosed patent claims. In general words, in a first aspect, a method of encoding multi-channel audio signals comprises generating of a first output signal, being encoding parameters representing a main signal. The main signal is a first linear combination of signals of at least a first and a second channel. The method further comprises generating of a second output signal, being encoding parameters representing a side signal. The side signal is a second linear combination of signals of at least the first and the second channel within an encoding frame. The method is characterised in that the generating of the second output signal further comprises scaling of the side signal to an energy contour of the main signal.
In a second aspect, a method of decoding multi-channel audio signals comprises generating of a decoded main signal from encoding parameters representing a main signal. The main signal is a first linear combination of signals of at least a first and a second channel. The method further comprises generating of a decoded side signal from encoding parameters representing a side signal. The side signal is a second linear combination of signals of at least a first and a second channel, within an encoding frame and is scaled to an energy contour of the main signal. The method further comprises combining of at least the decoded main signal and the decoded side signal into signals of at least the first and the second channel. The method is characterised in that the generating of a decoded side signal further comprises inverse scaling of the decoded side signal to an energy contour of the decoded main signal.
In a third aspect, an encoder apparatus comprises input means for multi-channel audio signals comprising at least a first and a second channel. The encoder apparatus comprises means for generating a first output signal, being encoding parameters representing a main signal. The main signal is a first linear combination of signals of at least the first and the second channel. The encoder apparatus further comprises means for generating a second output signal, being encoding parameters representing a side signal. The side signal is a second linear combination of signals of at least the first and the second channel, within an encoding frame. The encoder apparatus further comprises output means. The encoder apparatus is characterised in that the means for generating a second output signal further comprises means for scaling the side signal to an energy contour of the main signal.
In a fourth aspect, a decoder apparatus comprises input means for encoding parameters representing a main signal and encoding parameters representing a side signal. The main signal is a first linear combination of a first and a second channel. The side signal is a second linear combination of a first and a second channel and scaled to an energy contour of the main signal. The decoder apparatus further comprises means for generating a decoded main signal from the encoding parameters representing the main signal and means for generating a decoded side signal from the encoding parameters representing the side signal within an encoding frame. The decoder apparatus further comprises means for combining at least the decoded main signal and the decoded side signal into signals of at least a first and a second channel, and output means. The decoder apparatus is characterised in that the means for generating a decoded side signal in turn comprises means for inverse scaling the decoded side signal to an energy contour of the decoded main signal.
In a fifth aspect, an audio system comprises at least one of an encoder apparatus according to the third aspect and a decoder apparatus according to the fourth aspect.
The main advantage with the present invention is that the preservation of the perception of the audio signals is improved. Furthermore, the present invention still allows multi-channel signal transmission at very low bit rates.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention, together with further objects and advantages thereof, may best be understood by making reference to the following description taken together with the accompanying drawings, in which:

FIG. 1 is a block scheme of a system for transmitting polyphonic signals;
FIG. 2a is a block diagram of an encoder in a transmitter;
FIG. 2b is a block diagram of a decoder in a receiver;
FIG. 3a is a diagram illustrating encoding frames of different lengths;
FIGS. 3b and 3c are block diagrams of embodiments of side signal encoder units;
FIG. 4 is a block diagram of an embodiment of an encoder using balance factor encoding of side signal;
FIG. 5 is a block diagram of an embodiment of an encoder for multi-signal systems;
FIG. 6 is a block diagram of an embodiment of a decoder suitable for decoding signals from the device of Fig. 5;
FIG. 7a and b are diagrams illustrating a pre-echo artefact;
FIG. 8 is a block diagram of an embodiment of a side signal encoder unit, employing different encoding principles in different sub-frames;
FIG. 9 illustrates the use of different encoding principles in different frequency sub-bands;
FIG. 10 is a flow diagram of the basic steps of an embodiment of an encoding method; and
FIG. 11 is a flow diagram of the basic steps of an embodiment of a decoding method.

DETAILED DESCRIPTION

Fig. 1 illustrates a typical system 1, in which the present invention advantageously can be utilised. A transmitter 10 comprises an antenna 12 including associated hardware and software to be able to transmit radio signals 5 to a receiver 20. The transmitter 10 comprises among other parts a multi-channel encoder 14, which transforms signals of a number of input channels 16 into output signals suitable for radio transmission. Examples of suitable multi-channel encoders 14 are described in detail further below. The signals of the input channels 16 can be provided from e.g. an audio signal storage 18, such as a data file of digital representation of audio recordings, magnetic tape or vinyl disc recordings of audio etc. The signals of the input channels 16 can also be provided in "live", e.g. from a set of microphones 19. The audio signals are digitised, if not already in digital form, before entering the multi-channel encoder 14.
At the receiver 20 side, an antenna 22 with associated hardware and software handles the actual reception of radio signals 5 representing polyphonic audio signals. Here, typical functionalities, such as e.g. error correction, are performed. A decoder 24 decodes the received radio signals 5 and transforms the audio data carried thereby into signals of a number of output channels 26. The output signals can be provided to e.g. loudspeakers 29 for immediate presentation, or can be stored in an audio signal storage 28 of any kind.
The system 1 can for instance be a phone conference system, a system for supplying audio services or other audio applications. In some systems, such as e.g. the phone conference system, the communication has to be of a duplex type, while e.g. distribution of music from a service provider to a subscriber can be essentially of a one-way type. The transmission of signals from the transmitter 10 to the receiver 20 can also be performed by any other means, e.g. by different kinds of electromagnetic waves, cables or fibres as well as combinations thereof.
Fig. 2a illustrates an embodiment of an encoder. In this embodiment, the polyphonic signal is a stereo signal comprising two channels a and b, received at input 16A and 16B, respectively. The signals of channel a and b are provided to a pre-processing unit 32, where different signal conditioning procedures may be performed. The (perhaps modified) signals from the output of the pre-processing unit 32 are summed in an addition unit 34. This addition unit 34 also divides the sum by a factor of two. The signal x_mono produced in this way is a main signal of the stereo signals, since it basically comprises all data from both channels. In this embodiment the main signal thus represents a pure "mono" signal. The main signal x_mono is provided to a main signal encoder unit 38, which encodes the main signal according to any suitable encoding principles. Such principles are available within prior-art and are thus not further discussed here. The main signal encoder unit 38 gives an output signal p_mono, being encoding parameters representing a main signal.
In a subtraction unit 36, a difference (divided by a factor of two) of the channel signals is provided as a side signal x_side. In this embodiment, the side signal represents the difference between the two channels in the stereo signal. The side signal x_side is provided to a side signal encoding unit 30. Preferred embodiments of the side signal encoding unit 30 will be discussed further below. According to a side signal encoding procedure, which will be described more in detail further below, the side signal x_side is transferred into encoding parameters p_side representing a side signal x_side. In certain embodiments, this encoding takes place utilising also information of the main signal x_mono. The arrow 42 indicates such a provision, where the original uncoded main signal x_mono is utilised. In further other embodiments, the main signal information that is used in the side signal encoding unit 30 can be deduced from the encoding parameters p_mono representing the main signal, as indicated by the broken line 44.
The encoding parameters p_mono representing the main signal x_mono is a first output signal, and the encoding parameters p_side representing the side signal x_side is a second output signal. In a typical case, these two output signals p_mono, p_side, together representing the full stereo sound, are multiplexed into one transmission signal 52 in a multiplexor unit 40. However, in other embodiments, the transmission of the first and second output signals p_mono, p_side may take place separately.
In Fig. 2b, an embodiment of a decoder 24 is illustrated as a block scheme. The received signal 54, comprising encoding parameters representing the main and side signal information are provided to a demultiplexor unit 56, which separates a first and second input signal, respectively. The first input signal, corresponding to encoding parameters p_mono of a main signal, is provided to a main signal decoder unit 64. In a conventional manner, the encoding parameters p_mono representing the main signal are used to generate an decoded main signal x"_mono, being as similar to the main signal x_mono (Fig. 2a) of the encoder 14 (Fig. 2a) as possible.
Similarly, the second input signal, corresponding to a side signal, is provided to a side signal decoder unit 60. Here, the encoding parameters p_side representing the side signal are used to recover a decoded side signal x"_side. In some embodiments, the decoding procedure utilises information about the main signal x"_mono, as indicated by an arrow.
The decoded main and side signals x"_mono, x"_side are provided to an addition unit 70, which provides an output signal that is a representation of the original signal of channel a. Similarly, a difference provided by a subtraction unit 68 provides an output signal that is a representation of the original signal of channel b. These channel signals may be post-processed in a post-processor unit 74 according to prior-art signal processing procedures. Finally, the channel signals a and b are provided at the outputs 26A and 26B of the decoder.
As mentioned in the summary, encoding is typically performed in one frame at a time. A frame comprises audio samples within a pre-defined time period. In the bottom part of Fig. 3a, a frame SF2 of time duration L is illustrated. The audio samples within the unhatched portion are to be encoded together. The preceding samples and the subsequent samples are encoded in other frames. The division of the samples into frames will in any case introduce some discontinuities at the frame borders. Shifting sounds will give shifting encoding parameters, changing basically at each frame border. This will give rise to perceptible errors. One way to compensate somewhat for this is to base the encoding, not only on the samples that are to be encoded, but also on samples in the absolute vicinity of the frame, as indicated by the hatched portions. In such a way, there will be a softer transfer between the different frames. As an alternative, or complement, interpolation techniques are sometimes also utilised for reducing perception artefacts caused by frame borders. However, all such procedures require large additional computational resources, and for certain specific encoding techniques, it might also be difficult to provide it with any resources.
In this view, it is beneficial to utilise as long frames as possible, since the number of frame borders will be small. Also the coding efficiency typically becomes high and the necessary transmission bit-rate will typically be minimised. However, long frames give problems with pre-echo artefacts and ghost-like sounds.
By instead utilising shorter frames, such as SF1 or even SF0, having the durations of L/2 and L/4, respectively, anyone skilled in the art realises that the coding efficiency may be decreased, the transmission bit-rate may have to be higher and the problems with frame border artefacts will increase. However, shorter frames suffer less from e.g. other perception artefacts, such as ghost-like sounds and pre-echoing. In order to be able to minimise the coding error as much as possible, one should use an as short frame length as possible.
The audio perception will be improved by using a frame length for encoding of the side signal that is dependent on the present signal content. Since the influence of different frame lengths on the audio perception will differ depending on the nature of the sound to be encoded, an improvement can be obtained by letting the nature of the signal itself affect the frame length that is used. The encoding of the main signal is not the object of the present invention and is therefore not described in detail. However, the frame lengths used for the main signal may or may not be equal to the frame lengths used for the side signal.
Due to small temporal variations, it may e.g. in some cases be beneficial to encode the side signal with use of relatively long frames. This may be the case with recordings with a great amount of diffuse sound field such as concert recordings. In other cases, such as stereo speech conversation, short frames are probably to prefer. The decision which frame length is to prefer can be performed in two basic ways.
One embodiment of a side signal encoder unit 30 is illustrated in Fig. 3b, in which a closed loop decision is utilised. A basic encoding frame of length L is used here. A number of encoding schemes 81, characterised by a separate set 80 of sub-frames, are created. Each set 80 of sub-frames comprises one or more sub-frames of equal or differing lengths. The total length of the set 80 of sub-frames is, however, always equal to the basic encoding frame length L. With references to Fig. 3b, the top encoding scheme is characterised by a set of sub-frames comprising only one sub-frame of length L. The next set of sub-frames comprises two frames of length L/2. The third set comprises two frames of length L/4 followed by a L/2 frame.
The signal x_side provided to the side signal encoder unit 30 is encoded by all encoding schemes 81. In the top encoding scheme, the entire basic encoding frame is encoded in one piece. However, in the other encoding schemes, the signal x_side is encoded in each sub-frame separately from each other. The result from each encoding scheme is provided to a selector 85. A fidelity measurement means 83 determines a fidelity measure for each of the encoded signals. The fidelity measure is an objective quality value, preferably a signal-to-noise measure or a weighted signal-to-noise ratio. The fidelity measures associated with each encoding scheme are compared and the result controls a switching means 87 to select the encoding parameters representing the side signal from the encoding scheme giving the best fidelity measure as the output signal p_side from the side signal encoder unit 30.
Preferably, all possible combinations of frame lengths are tested and the set of sub-frames that gives the best objective quality, e.g. signal-to-noise ratio is selected.
In the present embodiment, the lengths of the sub-frames used are selected according to: $l_{sf} = {}^{l_{f}}{/_{2^{n}}},$

where l_sf are the lengths of the sub-frames, l_f is the length of the encoding frame and n is an integer. In the present embodiment, n is selected between 0 and 3. However, any frame lengths will be possible to use as long as the total length of the set is kept constant.
In Fig. 3c, another embodiment of a side signal encoder unit 30 is illustrated. Here, the frame length decision is an open loop decision, based on the statistics of the signal. In other words, the spectral characteristics of the side signal will be used as a base for deciding which encoding scheme that is going to be used. As before, different encoding schemes characterised by different sets of sub-frames are available. However, in this embodiment, the selector 85 is placed before the actual encoding. The input side signal x_side enters the selector 85 and a signal analysing unit 84. The result of the analysis becomes the input of a switch 86, in which only one of the encoding schemes 81 are utilised. The output from that encoding scheme will also be the output signal p_side from the side signal encoder unit 30.
The advantage with an open loop decision is that only one actual encoding has to be performed. The disadvantage is, however, that the analysis of the signal characteristics may be very complicated indeed and it may be difficult to predict possible behaviours in advance to be able to give an appropriate choice in the switch 86. A lot of statistical analysis of sound has to be performed and included in the signal analysing unit 84. Any small change in the encoding schemes may turn upside down on the statistical behaviour.
By using closed loop selection (Fig. 3b), encoding schemes may be exchanged without making any changes in the rest of the unit. On the other hand, if many encoding schemes are to be investigated, the computational requirements will be high.
The benefit with such a variable frame length coding for the side signal is that one can select between a fine temporal resolution and coarse frequency resolution on one side and coarse temporal resolution and fine frequency resolution on the other. The above embodiments will preserve the stereo image in the best possible manner.
There are also some requirements on the actual encoding utilised in the different encoding schemes. In particular when the closed loop selection is used, the computational resources to perform a number of more or less simultaneous encoding have to be large. The more complicated the encoding process is, the more computational power is needed. Furthermore, a low bit rate at transmission is also to prefer.
The method presented in US 5,434,948 , uses a filtered version of the mono (main) signal to resemble the side or difference signal. The filter parameters are optimised and allowed to vary in time. The filter parameters are then transmitted representing an encoding of the side signal. In one embodiment, also a residual side signal is transmitted. In many cases, such an approach would be possible to use as side signal encoding method. This approach has, however, some disadvantages. The quantisation of the filter coefficients and any residual side signal often require relatively high bit rates for transmission, since the filter order has to be high to provide an accurate side signal estimate. The estimation of the filter itself may be problematic, especially in cases of transient rich music. Estimation errors will give a modified side signal that is sometimes larger in magnitude than the unmodified signal. This will lead to higher bit rate demands. Moreover, if a new set of filter coefficients are computed every N samples, the filter coefficients need to be interpolated to yield a smooth transition from one set of filter coefficients to another, as discussed above. Interpolation of filter coefficients is a complex task and errors in the interpolation will manifest itself in large side error signals leading to higher bit rates needed for the difference error signal encoder.
A means to avoid the need for interpolation is to update the filter coefficients on a sample-by-sample basis and rely on backwards-adaptive analysis. For this to work well it is needed that the bit rate of the residual encoder is fairly high. This is therefore not a good alternative for low bit rate stereo coding.
There exist cases, e.g. quite common with music, where the mono and the difference signals are almost un-correlated. The filter estimation then becomes very troublesome with the added risk of just making things worse for the difference error signal encoder.
The solution according to US 5,434,948 can work pretty well in cases where the filter coefficients vary very slowly in time, e.g. conference telephony systems. In the case of music signals, this approach does not work very well as the filters need to change very fast to track the stereo image. This means that sub-frame lengths of very differing magnitude has to be utilised, which means that the number of combinations to test increases rapidly. This in turn means that the requirements for computing all possible encoding schemes becomes impracticably high.
Therefore, in a preferred embodiment, the encoding of the side signal is based on the idea to reduce the redundancy between the mono and side signal by using a simple balance factor instead of a complex bit rate consuming predictor filter. The residual of this operation is then encoded. The magnitude of such a residual is relatively small and does not call for very high bit rate need for transfer. This idea is very suitable indeed to combine with the variable frame set approach described earlier, since the computational complexity is low.
The use of a balance factor combined with the variable frame length approach removes the need for complex interpolation and the associated problems that interpolation may cause. Moreover, the use of a simple balance factor instead of a complex filter gives fewer problems with estimation as possible estimation errors for the balance factor has less impact. The preferred solution will be able to reproduce both panned signals and diffuse sound fields with good quality and with limited bit rate requirements and computational resources.
Fig. 4 illustrates a preferred embodiment of a stereo encoder. This embodiment is very similar to the one shown in Fig. 2a, however, with the details of the side signal encoder unit 30 revealed. The encoder 14 of this embodiment does not have any pre-processing unit, and the input signals are provided directly to the addition and subtraction units 34, 36. The mono signal x_mono is multiplied with a certain balance factor g_sm in a multiplier 33. In a subtraction unit 35, the multiplied mono signal is subtracted from the side signal x_side, i.e. essentially the difference between the two channels, to produce a side residual signal. The balance factor g_sm is determined based on the content of the mono and side signals by the optimiser 37 in order to minimise the side residual signal according to a quality criterion. The quality criterion is preferably a least mean square criterion. The side residual signal is encoded in a side residual encoder 39 according to any encoder procedures. Preferably, the side residual encoder 39 is a low bit rate transform encoder or a CELP (Codebook Excited Linear Prediction) encoder. The encoding parameters p_side representing the side signal then comprises the encoding parameters p_{side residual} representing the side residual signal and the optimised balance factor 49.
In the embodiment of Fig. 4, the mono signal 42 used for synthesising the side signals is the target signal x_mono for the mono encoder 38. As mentioned above (in connection with Fig. 2a), the local synthesis signal of the mono encoder 38 can also be utilised. In the latter case, the total encoder delay may be increased and the computational complexity for the side signal may increase. On the other hand, the quality may be better as it is then possible to repair coding errors made in the mono encoder.
In a more mathematical way, the basic encoding scheme can be described as follows. Denote the two channel signals as a and b, which may be the left and right channel of a stereo pair. The channel signals are combined into a mono signal by addition and to a side signal by a subtraction. In equation form, the operations are described as: $x_{mono} (n) = 0.5 (a (n) + b (n))$
$x_{side} (n) = 0.5 (a (n) - b (n)) .$
It is beneficial to scale the x_mono and x_side signals down by a factor of two. It is here implied that other ways of creating the x_mono and x_side exist. One can for instance use: $x_{mono} (n) = γa (n) + (1 - γ) b (n)$
$x_{side} (n) = γa (n) - (1 - γ) b (n)$
$0 \leq γ \leq 1.0.$
On blocks of the input signals, a modified or residual side signal is computed according to: $x_{side residual} (n) = x_{side} (n) - f (x_{mono} x_{side}) x_{mono} (n),$

where f(x_mono,x_side) is a balance factor function that based on the block on N samples, i.e. a sub-frame, from the side and mono signals strive to remove as much as possible from the side signal. In other words, the balance factor is used to minimise the residual side signal. In the special case where it is minimised in a mean square sense, this is equivalent to minimising the energy of the residual side signal x_{side residual}.
In the above mentioned special case f(x_mono,x_side ) is described as: $f (x_{mono} x_{side}) = \frac{R_{sm}}{R_{mm}}$
$R_{mm} = [\sum_{n = frame start}^{frame end} x_{mono} (n) x_{mono} (n)]$
$R_{sm} = [\sum_{n = frame start}^{frame end} x_{side} (n) x_{mono} (n)],$

where x_side is the side signal and x_mono is the mono signal. Note that the function is based on a block starting at "frame start" and ending at "frame end".
It is possible to add weighting in the frequency domain to the computation of the balance factor. This is done by convoluting the x_side and x_mono signals with the impulse response of a weighting filter. It is then possible to move the estimation error to a frequency range where they are less easy to hear. This is referred to as perceptual weighting.
A quantized version of the balance factor value given by the function f(x_mono,x_side ) is transmitted to the decoder. It is preferable to account for the quantization already when the modified side signal is generated. The expression below is then achieved: $x_{side residual} (n) = x_{side} (n) - g_{Q} x_{mono} (n)$
$g_{Q} = Q_{g}^{- 1} (Q_{g} (\frac{R_{sm}}{R_{mm}})) .$
Q_g (..) is a quantization function that is applied to the balance factor given by the function f(x_mono,x_side). The balance factor is transmitted on the transmission channel. In normal left-right panned signals the balance factor is limited to the interval [-1.0 1.0]. If on the other hand the channels are out of phase with regards to one another, the balance factor may extend beyond these limits.
As an optional means to stabilise the stereo image, one can limit the balance factor if the normalised cross correlation between the mono and the side signal is poor as given by the equation below: $g_{Q} = Q_{g}^{- 1} (Q_{g} (|\underset{= sm}{R}| \frac{R_{sm}}{R_{mm}})),$

where $\underset{= sm}{R} = \frac{R_{sm}}{\sqrt{R_{ss} \cdot R_{mm}}}$
$R_{sm} = [\sum_{n = frame start}^{frame end} x_{side} (n) x_{mono} (n)] .$
These situations occur quite frequently with e.g. classical music or studio music with a great amount of diffuse sounds, where in some cases the a and b channels might almost cancel out one another on occasions when a mono signal is created. The effect on the balance factor is that is can jump rapidly, causing a confused stereo image. The fix above alleviates this problem.
The filter-based approach in US 5,434,948 has the similar problems, but in that case the solution is not so simple.
If E_s is the encoding function (e.g. a transform encoder) of the residual side signal and E_m is the encoding function of the mono signal, then the decoded a" and b" signals in the decoder end can be described as (it is assumed here that γ = 0.5). $aʺ (n) = (1 + g_{Q}) x_{mono}^{ʺ} (n) + x_{side}^{ʺ} (n)$
$bʺ (n) = (1 - g_{Q}) x_{mono}^{ʺ} (n) + x_{side}^{ʺ} (n)$
$x_{side}^{ʺ} = E_{s}^{- 1} (E_{s} (x_{side residual}))$
$x_{mono}^{ʺ} = E_{m}^{- 1} (E_{m} (x_{mono}))$
One important benefit from computing the balance factor for each frame is that one avoids the use of interpolation. Instead, normally, as described above, the frame processing is performed with overlapping frames.
The encoding principle using balance factors operates particularly well in the case of music signals, where fast changes typically are needed to track the stereo image.
Lately, multi-channel coding has become popular. One example is 5.1 channel surround sound in DVD movies. The channels are there arranged as: front left, front centre, front right, rear left, rear right and subwoofer. In Fig. 5, an embodiment of an encoder that encodes the three front channels in such an arrangement exploiting interchannel redundancies is shown.
Three channel signals L, C, R are provided on three inputs 16A-C, and the mono signal x_mono is created by a sum of all three signals. A centre signal encoder unit 130 is added, which receives the centre signal x_centre. The mono signal 42 is in this embodiment the encoded and decoded mono signal x"_mono, and is multiplied with a certain balance factor g_Q in a multiplier 133. In a subtraction unit 135, the multiplied mono signal is subtracted from the centre signal x_centre, to produce a centre residual signal. The balance factor g_Q is determined based on the content of the mono and centre signals by an optimiser 137 in order to minimise the centre residual signal according to the quality criterion. The centre residual signal is encoded in a centre residual encoder 139 according to any encoder procedures. Preferably, the centre residual encoder 139 is a low bit rate transform encoder or a CELP encoder. The encoding parameters p_centre representing the centre signal then comprises the encoding parameters p_{centre residual} representing the centre residual signal and the optimised balance factor 149. The centre residual signal and the scaled mono signal are added in an addition unit 235, creating a modified centre signal 142 being compensated for encoding errors.
The side signal x_side, i.e. the difference between the left L and right R channels is provided to the side signal encoder unit 30 as in earlier embodiments. However, here, the optimiser 37 also depends on the modified centre signal 142 provided by the centre signal encoder unit 130. The side residual signal will therefore be created as an optimum linear combination of the mono signal 42, the modified centre signal 142 and the side signal in the subtraction unit 35.
The variable frame length concept described above can be applied on either of the side and centre signals, or on both.
Fig. 6 illustrates a decoder unit suitable for receiving encoded audio signals from the encoder unit of Fig. 5. The received signal 54 is divided into encoding parameters p_mono representing the main signal, encoding parameters p_centre representing the centre signal and encoding parameters p_side representing the side signal. In the decoder 64, the encoding parameters p_mono representing the main signal are used to generate a main signal x"_mono. In the decoder 160, the encoding parameters p_centre representing the centre signal are used to generate a centre signal x"_centre, based on main signal x"_mono. In the decoder 60, the encoding parameters p_side representing the side signal are decoded, generating a side signal x"_side, based on main signal x"_mono and centre signal x"_centre.
The procedure can be mathematically expressed as follows:
The input signals x_left, x_right and x_centre are combined to a mono channel according to: $x_{mono} (n) = {ax}_{left} (n) + {βx}_{right} (n) + {χx}_{centre} (n) .$
α, β and χ are in the remaining section set to 1.0 for simplicity, but they can be set to arbitrary values. The α, β and χ values can be either constant or dependent of the signal contents in order to emphasise one or two channels in order to achieve an optimal quality.
The normalised cross correlation between the mono and the centre signal is computed as: $\underset{= cm}{R} = \frac{R_{cm}}{\sqrt{R_{cc} \cdot R_{mm}}},$

where $R_{cc} = [\sum_{n = frame start}^{frame end} x_{centre} (n) x_{centre} (n)]$
$R_{mm} = [\sum_{n = frame start}^{frame end} x_{mono} (n) x_{mono} (n)]$
$R_{cm} = [\sum_{n = frame start}^{frame end} x_{centre} (n) x_{mono} (n)] .$
x_centre is the centre signal and x_mono is the mono signal. The mono signal comes from the mono target signal but it is possible to use the local synthesis of the mono encoder as well.
The centre residual signal to be encoded is: $x_{centre residual} (n) = x_{centre} (n) - g_{Q} x_{mono} (n)$
$g_{Q} = Q_{g}^{- 1} (Q_{g} (\frac{R_{cm}}{R_{mm}})) .$
Q_g (..) is a quantization function that is applied to the balance factor. The balance factor is transmitted on the transmission channel.
If E_c is the encoding function (e.g. a transform encoder) of the centre residual signal and E_m is the encoding function of the mono signal then the decoded x"_centre signal in the decoder end can be described as: $x_{centre}^{ʺ} (n) = g_{Q} x_{mono}^{ʺ} (n) + x_{centre residual}^{ʺ} (n)$
$x_{centre residual}^{ʺ} = E_{c}^{- 1} (E_{c} (x_{centre residual}))$
$x_{mono}^{ʺ} = E_{m}^{- 1} (E_{m} (x_{mono}))$
The side residual signal to be encoded is: $x_{side residual} (n) = (x_{left} (n) {- x}_{side} (n)) - g_{Qsm} x_{mono}^{ʺ} (n) - g_{Qsc} x_{mono}^{ʺ} (n),$

where g_Qsm and g_Qsc are quantized values of the parameters g_sm and g_sc that minimises the expression: $\sum_{n = frame start}^{frame end} {[| (x_{left} (n) {- x}_{right} (n)) - g_{sm} x_{mono}^{ʺ} (n) - g_{sc} x_{centre}^{ʺ} (n) |]}^{η} .$
η can for instance be equal to 2 for a least square minimisation of the error. The g_sm and g_sc parameters can be quantized jointly or separately.
If E_s is the encoding function of the side residual signal, then the decoded x"_left and x"_right channel signals are given as: $x_{left}^{ʺ} (n) = x_{mono}^{ʺ} (n) - x_{centre}^{ʺ} (n) + x_{side}^{ʺ} (n)$
$x_{right}^{ʺ} (n) = x_{mono}^{ʺ} (n) - x_{centre}^{ʺ} (n) - x_{side}^{ʺ} (n)$
$x_{side}^{ʺ} (n) = x_{side redidual}^{ʺ} + g_{Qsm} x_{mono}^{ʺ} (n) + g_{Qsc} x_{centre}^{ʺ} (n)$
$x_{side residual}^{ʺ} = E_{s}^{- 1} (E_{s} (x_{side residual})) .$
One of the perception artefacts that are most annoying is the pre-echo effect. In Fig. 7a-b, diagrams are illustrating such an artefact. Assume a signal component having the time development as shown by curve 100. In the beginning, starting from t0, the signal component is not present in the audio sample. At a time t between t1 and t2, the signal component suddenly appears. When the signal component is encoded, using a frame length of t2-t1, the occurrence of the signal component will be "smeared out" over the entire frame, as indicated in curve 101. If a decoding takes place of the curve 101, the signal component appears a time Δt before the intended appearance of the signal component, and a "pre-echo" is perceived.
The pre-echoing artefacts become more accentuated if long encoding frames are used. By using shorter frames, the artefact is somewhat suppressed. Another way to deal with the pre-echoing problems described above is to utilise the fact that the mono signal is available at both the encoder and decoder end. This makes it possible to scale the side signal according to the energy contour of the mono signal. In the decoder end, the inverse scaling is performed and thus some of the pre-echo problems may be alleviated.
An energy contour of the mono signal is computed over the frame as: $E_{c} (m) = [\sum_{n = m - L}^{m + L} w (n) x_{mono}^{2} (n)], frame start \leq m \leq frame end,$

where w(n) is a windowing function. The simplest windowing function is a rectangular window, but other window types such as a hamming window may be more desirable.
The side residual signal is then scaled as: ${\underset{̲}{x}}_{side residual} (n) = \frac{x_{side residual} (n)}{\sqrt{E_{c} (n)}}, frame start \leq n \leq frame end .$
In a more general form the equation above can be written as: ${\underset{̲}{x}}_{side residual} (n) = \frac{x_{side residual} (n)}{f (E_{c} (n))}, frame start \leq n \leq frame end,$

where f(..) is a monotonic continuous function. In the decoder, the energy contour is computed on the decoded mono signal and is applied to the decoded side signal as: ${\underset{̲}{x}}_{side}^{ʺ} (n) = x_{side}^{ʺ} (n) f (E_{c} (n)), frame start \leq n \leq frame end .$
Since this energy contour scaling in some sense is alternative to the use of shorter frame lengths, this concept is particularly well suited to be combined with the variable frame length concept, described further above. By having some encoding schemes that applies energy contour scaling, some that do not and some that applies energy contour scaling only during certain sub-frames, a more flexible set of encoding schemes may be provided. In Fig. 8, an embodiment of a signal encoder unit 30 according to the present invention is illustrated. Here, the different encoding schemes 81 comprise hatched sub-frames, representing encoding applying the energy contour scaling, and un-hatched sub-frames, representing encoding procedures not applying the energy contour scaling. In this manner, combinations not only of sub-frames of differing lengths, but sub-frames also of differing encoding principles are available. In the present explanatory example, the application of energy contour scaling differs between different encoding schemes. In a more general case, any encoding principles can be combined with the variable length concept in an analogous manner.
The set of encoding schemes of Fig. 8 comprises schemes that handle e.g. pre-echoing artefacts in different ways. In some schemes, longer sub-frames with pre-echoing minimisation according to the energy contour principle are used. In other schemes, shorter sub-frames without energy contour scaling are utilised. Depending on the signal content, one of the alternatives may be more advantageous. For very severe pre-echoing cases, encoding schemes utilising short sub-frames with energy contour scaling may be necessary.
The proposed solution can be used in the full frequency band or in one or more distinct sub bands. The use of sub-band can be applied either on both the main and side signals, or on one of them separately. A preferred embodiment comprises a split of the side signal in several frequency bands. The reason is simply that it is easier to remove the possible redundancy in an isolated frequency band than in the entire frequency band. This is particularly important when encoding music signals with rich spectral content.
One possible use is to encode the frequency band below a pre-determined threshold with the above method. The pre-determined threshold can preferably be 2 kHz, or even more preferably 1 kHz. For the remaining part of the frequency range of interest, one can either encode another additional frequency band with the above method, or use a completely different method.
One motivation to use the above method preferably for low frequencies is that the diffuse sound fields generally have little energy content at high frequencies. The natural reason is that sound absorption typically increases with frequency. Also, the diffuse sound field components seem to play a less important role for the human auditory system at higher frequencies.
Therefore, it is beneficial to employ this solution at low frequencies (below 1 or 2 kHz) and rely on other, even more bit efficient coding schemes at higher frequencies. The fact that the scheme is only applied at low frequencies gives a large saving in bit rate as the necessary bit rate with the proposed method is proportional to the required bandwidth. In most cases, the mono encoder can encode the entire frequency band, while the proposed side signal encoding is suggested to be performed only in the lower part of the frequency band, as schematically illustrated by Fig. 9. Reference number 301 refers to an encoding scheme of the side signal, reference number 302 refers to any other encoding scheme of the side signal and reference number 303 refers to an encoding scheme of the side signal.
There also exist the possibility to use the proposed method for several distinct frequency bands.
In Fig. 10, the main steps of an embodiment of an encoding method are illustrated as a flow diagram. The procedure starts in step 200. In step 210, a main signal deduced from the polyphonic signals is encoded. In step 212, encoding schemes are provided, which comprise sub-frames with differing lengths and/or order. A side signal deduced in step 214 from the polyphonic signals is encoded by an encoding scheme selected dependent at least partly on the actual signal content of the present polyphonic signals. The procedure ends in step 299.
In Fig. 11, the main steps of an embodiment of a decoding method are illustrated as a flow diagram. The procedure starts in step 200. In step 220, a received encoded main signal is decoded. In step 222, encoding schemes are provided, which comprise sub-frames with differing lengths and/or order. A received side signal is decoded in step 224 by a selected encoding scheme. In step 226, the decoded main and side signals are combined to a polyphonic signal. The procedure ends in step 299.
The embodiments described above are to be understood as a few illustrative examples of the present invention. It will be understood by those skilled in the art that various modifications, combinations and changes may be made to the embodiments without departing from the scope of the present invention. In particular, different part solutions in the different embodiments can be combined in other configurations, where technically possible. The scope of the present invention is, however, defined by the appended claims.

REFERENCES

European patent 0497413
US patent 5,285,498
US patent 5,434,948
"Binaural cue coding applied to stereo and multi-channel audio compression", 112th AES convention, May 2002, Munich, Germany by C. Faller et al.

Claims

A method of encoding multi-channel audio signals, comprising the steps of:
generating (210) a first output signal (p_mono) being encoding parameters representing a main signal (x_mono);

said main signal (x_mono) being a first linear combination of signals of at least a first and a second channel (a, b; L, R); and

generating (214) a second output signal (p_side) being encoding parameters representing a side signal (x_side);

said side (x_side) signal being a second linear combination of signals of at least the first and the second channel (a, b; L, R) within an encoding frame (80),
characterised in that the step of generating the second output signal (p_side) further comprises the step of:
scaling the side signal (x_side) to an energy contour of the main signal (x_mono).
A method according to claim 1, characterised in that the side signal (x_side) is scaled by a factor being a monotonic continuous function of the energy contour of the main signal (x_mono).
A method according to claim 1, characterised in that the step of generating a second output signal (p_side) comprises the step of creating a side residual signal (x_{side residual}) based on a balanced difference between the side signal (x_side) and the main signal (x_mono), whereby the residual side signal (x_{side residual}) is scaled to an energy contour of the main signal (x_mono).
A method according to claim 3, characterised in that the side residual signal (x_{side residual}) is divided by a factor being a monotonic continuous function of the energy contour of the main signal (x_mono).
A method of decoding multi-channel audio signals, comprising the steps of:
generating (220) a decoded main signal (x"_mono) from encoding parameters (p_mono) representing a main signal (x_mono);

said main signal (x_mono) being a first linear combination of signals of at least a first and a second channel (a, b; L, R);

generating (224) a decoded side signal (x"_side) from encoding parameters (p_side) representing a side signal (x_side);

said side signal (x_side) being a second linear combination of signals of at least the first and the second channel (a, b; L, R), within an encoding frame (80); and

combining (226) at least the decoded main signal (x"_mono) and the decoded side signal (x"_side) into signals of at least said first and said second channel (a, b; L, R),
characterised in that
said side signal (x_side) being scaled to an energy contour of the main signal (x_mono);
the step of generating the decoded side signal (x"_side) further comprises the step of:
inverse scaling the decoded side signal (x"_side) to an energy contour of the decoded main signal (x"_mono).
A method according to claim 5, characterised in that the decoded side signal (x"_side) is inverse scaled by a factor being a monotonic continuous function of the energy contour of the decoded main signal (x"_mono).
A method according to claim 5, characterised in that the step of generating (224) the decoded side signal (x"_side) comprises the step of generating a decoded side residual signal (x"_{side residual}) and generating the decoded side signal (x"_side) based on the decoded side residual signal (x"_{side residual}), whereby the decoded side residual signal (x"_{side residual}) is inverse scaled to an energy contour of the decoded main signal (x"_mono).
A method according to claim 7, characterised in that the decoded side residual signal (x"_{side residual}) is multiplied by a factor being a monotonic continuous function of the energy contour of the decoded main signal (x"_mono).
Encoder apparatus (14), comprising:
input means (16; 16A-C) for multi-channel audio signals (a, b; L, R, C) comprising at least a first and a second channel (a, b; L, R),

means (38) for generating a first output signal (p_mono) being encoding parameters representing a main signal (x_mono);

said main signal (x_mono) being a first linear combination of signals of at least the first and the second channel (a, b; L, R);

means (30) for generating a second output signal (p_side) being encoding parameters representing a side signal (x_side);

said side signal (x_side) being a second linear combination of signals of at least the first and the second channel (a, b; L, R), within an encoding frame (80); and

output means (52);
characterised in that the means for generating a second output signal (p_side) further comprises:
means for scaling the side signal (x_side) to an energy contour of the main signal (x_mono).
Encoder apparatus according to claim 9, characterised in that the means for scaling the side signal (x_side) are adapted to scale the side signal (x_side) by a factor being a monotonic continuous function of the energy contour of the main signal (x_mono).
Encoder apparatus according to claim 9, characterised in that the means for generating the second output signal (p_side) further comprises means for creating a side residual signal (X_{side residual}) based on a balanced difference between the side signal (x_side) and the main signal (x_mono), whereby the means for scaling the side signal (x_side) are adapted to scale the side residual signal (x_{side residual}) to an energy contour of the main signal (x_mono).
Encoder apparatus according to claim 11, characterised in that the means for scaling the side signal (x_side) are adapted to divide the side residual signal (x_{side residual}) by a factor being a monotonic continuous function of the energy contour of the main signal (x_mono).
Decoder apparatus (24), comprising:
input means (54) for encoding parameters (p_mono) representing a main signal and encoding parameters (p_side) representing a side signal;

said main signal (x_mono) being a first linear combination of a first and a second channel (a, b; L, R);

said side signal (x_side) being a second linear combination of the first and the second channel (a, b; L, R);

means (64) for generating a decoded main signal (x"_mono) from the encoding parameters (p_mono) representing the main signal;

means (60) for generating a decoded side signal (x"_side) from the encoding parameters (p_side) representing the side signal within an encoding frame (80);

means (68, 70) for combining at least the decoded main signal (x"_mono) and the decoded side signal (x"_side) into signals of at least the first and the second channel (a, b; L, R); and

output means (26; 26A-C),
characterised in that
said side signal (x_side) being scaled to an energy contour of the main signal (x_mono);
the means (60) for generating the decoded side signal (x"_side) in turn comprises:
means for inverse scaling the decoded side signal (x"_side) to an energy contour of the decoded main signal (x"_mono).
Decoder apparatus according to claim 13, characterised in that the means for inverse scaling the decoded side signal (x"_side) are adapted to inverse scale the decoded side signal (x"_side) by a factor being a monotonic continuous function of the energy contour of the decoded main signal (x"_mono).
Decoder apparatus according to claim 13, characterised in that the means (60) for generating the decoded side signal (x"_side) further comprises means for generating a decoded side residual signal (x"_{side residual}) and for generating the decoded side signal (x"_side) based on the decoded side residual signal (x"_{side residual}), whereby the means for inverse scaling the decoded side signal (x"_side) are adapted to inverse scale the decoded side residual signal (x_{side residual}) to an energy contour of the decoded main signal (x"_mono).
Decoder apparatus according to claim 15, characterised in that the means for inverse scaling the decoded side signal (x"_side) are adapted to multiply the decoded side residual signal (x_{side residual}) by a factor being a monotonic continuous function of the energy contour of the decoded main signal (x"_mono).
Audio system (1) comprising at least one of:
an encoder apparatus (14) according to any of the claims 9 to 12, and

a decoder apparatus (24) according to any of the claims 13 to 16.