1 MATRIX ENCODER WITH IMPROVED CHANNEL SEPARATION The present application is a divisional application from Australian Patent Application No. 2012267193, the entire disclosure of which is incorporated herein by reference. FIELD OF THE INVENTION The present invention relates to an improved matrix encoder for surround sound. The matrix encoder may be associated with a surround sound system wherein at least four audio input signals representing an original sound field are encoded into two channels and the two channels are decoded into at least four channels corresponding to the four audio input signals. CROSS REFERENCE TO RELATED APPLICATION The present invention is related to the following international patent application assigned to the present applicant the disclosure of which is incorporated herein by cross reference: PCT/AU201 0/001 666 - IMPROVED MATRIX DECODER FOR SURROUND SOUND BACKGROUND OF THE INVENTION In a multi-channel system as described above four channels of audio signals are obtained from an original sound field and are encoded by an encoder into two channels. The encoded two channels may be recorded on recording media such as CD. DVD or the like or broadcast via stereo TV or FM radio. The encoded two channels may be reproduced from the recording media or broadcast and decoded by means of a matrix decoder back into four channels approximating the four channels of audio signals obtained from the original sound field. The decoded signals may be applied to four speakers to reproduce the original sound field through suitable amplifiers. To facilitate an understanding of the present invention the principles of a 4-2-4 matrix playback system and a conventional encoder is described below with reference to FIGS. 1 and 2 of the accompanying drawings. In the system shown in FIG. 1, four microphones 10, 11, 12 and 13 are installed in an original sound field 14 in order to produce four channel audio signals FL (front-left), FR (front-right), RL (rear-left) and RR (rear-right) respectively. An optional centre channel may also be produced. The four channel audio signals are supplied to 2 encoder 15 to be transformed or encoded into two signals L and R. The outputs L and R from encoder 15 are applied to a decoder 16 to be transformed or decoded into reproduced four channel signals FL, FR', RL' and RR' approximating the original four channel signals FL, FR, RL and RR. Decoder 16 may include single or multi-band 5 processing as described below, The reproduced four channel signals may be applied through amplifiers (not shown) to four loud speakers 17, 18, 19 and 20 located in a listening space 21 to provide a multichannel sound field that more closely approximates the original sound field 14 when compared to a prior art two channel system. 10 A variety of two channel systems 22 including CD, DVD TV, FM radio, etc. may be used to capture or store outputs L and R from encoder 15 and to supply the captured or stored outputs to decoder 16. In one example outputs L and R from encoder 15 may be recorded on a storage medium such as a CD, DVD or magnetic tape and the 15 outputs from the storage medium may be applied to decoder 16. According to another example the outputs L and R from encoder 15 or the outputs reproduced from the recording medium may be transmitted to decoder 16 via a stereo TV or an FM stereo radio broadcasting system 20 Examples of a conventional encoder 15 include Q sound, Prologic or conventional stereo Encoder 15 in FIG, I may be configured as shown in FIG 2 wherein audio signals FL and FR produced by microphones 10 and 11 disposed in the front of original sound field 14, and audio signals RL and RR produced by microphones 12. 13 disposed in the rear of original sound field 14 are applied to a conventional matrix 25 circuit 23 Matrix circuit 23 includes a plurality of addersmultipliers and phase shifters arranged to produce L and R output signals as follows: 30 L =FL + kFR + jRL +J kRR R = FR + kFL -jRR -jkRL wherein k denotes a transformation or matrix coefficient generally having a value approximately 0.41 4 and j denotes a 90 degree phase shift. The phase shifters may 3 provide a substantially consistent phase shift over the entire audio frequency band. The four channel signals FL, FR, RL' and RRr may be reproduced by a conventional decoder having the same matrix coefficient k. it may be shown that when matrix coefficient k = 0.41 4, separations between channel FL' and adjacent channels FR' 5 and RU are respectively equal to -3dB and separation between the channels FL and RR' in a diagonal direction equals -infin. dB. Because the separation between adjacent channels equals -3dB it is not possible to enjoy stereo playback of four channels with a sufficiently large directional resolution. 10 FG. 3 shows a block diagram of a decoder including a variable matrix 24 having control unit 25 and decoder unit 26 and employing matrix coefficients SL SR, SF. SB the magnitudes of which may be controlled in accordance with the phase difference between two channel signals L and P 15 in the decoder shown in FIG, 3, the two channel signals L and R are applied to input terminaIs 27 and 28 of the decoder from a two-channel media source and hence to input terminals 29 and 30 of variable matrix 24. Input terminals 27 and 28 are also coupled to input terminals 3 i and 32 of variable matrix 24 via 90 degree phase shift circuit 33, Variable matrix 24 operates to decode or dernatrix the two channel signals 20 L aRd R to produce four channel signals at its output terminals 34, 35, 36 and 37, Control unit 25 provides steering control signals SL, SR, SF, and SB to decoder unit 26 in accordance with the phase difference between two-channel signals L and R, The magnitudes of the steering control signals SL, SR, SF, and SB from control unit 25 may vary in opposite directions in proportion to the phase difference between 25 signals L and R Control signal SF may be used to control the matrix coefficient related to the front channels and control signal SB may be used to control the matrix coefficient related to the rear channels. Similarly control signal SR may be used to control the matrix coefficient related to the right channels and control signal SL may be used to control the matrix coefficient related to the left channels, Where the phase 30 difference between signals L and R is near zero, for instance, the control signal SF operates to decrease the matrix coefficient related to the front channels thus enhancing separation between the front channels. On the other hand, control signal SB operates to increase the matrix coefficient related to the rear channels to reduce separation between rear channels. Concurrently therewith signal levels of the front 4 channels may be increased and those of the rear channels may be decreased to improve separation between the front and rear channels. The control unit 25 may include a phase discriminator for detecting a phase difference 5 between signals L and R or a comparator for detecting a phase relationship between signals L and R in terms of the difference in the levels of a sum signal (L+R) and a difference signal (L-R) A reason for controlling the matrix coefficient associated with the front and rear channels by detecting the phase relationship between signals L and R is that humans have a keen sensitivity to detect the direction of a large sound but 10 sensitivity for a small sound coexisting with the large sound may be relatively poor. Consequently, where there is a large sound in the front and a small sound in the rear playback of four channels may be more efficient if separation between the front channels is enhanced and separation between the rear channels is reduced. in contrast, where a small sound exists in the front and a large sound in the rear 15 playback of four channels may be more efficient if separation between the rear channels is enhanced and separation between the front channels is reduced Where a large sound is present in the front and a small sound is present in the rear, that is., where FL, FR w RL, RR, signals L and R may have substantially the same 20 phase. This means that the level of a sum signal (L + R) may be higher than that of a difference signal (L PR Conversely, where a large sound is present in the rear while a small sound is present in the front that is, where FL, FR < RL R. signals L and R have opposite phase. In 25 such a case, the level of the sum signal (L + R) may be lower than the level of the difference signal (L - R). For this reason, it may be possible to detect phase relationship between signals L and R by either a phase discriminator or a comparator. A variable matrix decoder is described in international patent application 30 PCT/AU201 01001 666 assigned to the present applicant The decoder with its intelligent tri band steering systems may achieve approximately 40 db channel separation between all decoded surround outputs on dynamic music content. One disadvantage of the decoder is that stereo encoded media lacks full left/right channel separation and sounds somewhat narrowed, 5 in pre digital (CD) days it was commonly accepted that 20 d b separation was desirable so no crosstalk could be heard. However up to 100db separation is achievable with modern digital technology. Nevertheless the question still persists as 5 to what level of separation is acceptable to be undetectable in practical terms by human hearing under typical music conditions. Contrary to common belief, the direction from which sound arrives is perceived by the human ear based on both arrival time and loudness, not loudness alone, This is a 10 psychoacoustic phenomenon known as the "HAAS" or "precedence" effect and is illustrated by a curve as shown in Figure 4. For wave fronts with arrival time differences in a range of 1-30 milliseconds, and sound pressure level differences of up to 12 db, arrival time is the dominant determinant of perceived sound direction This is the region underneath the curve. Hence sound is perceived as coming from 15 the direction of a first wave front to arrive, even if the first wave front may be up to 12 db lower in sound pressure level than a later wave front. The Haas curve basically suggests that 12 db signal level difference is required to overcome time delay clues of left/right image positioning. When a separation of 12 db was tested compared to the 100 db available with modern CD technology it was found that listeners could not pick 20 any difference. When the encoder shown in Fig. 2 is used there is an excess of surround separation amounting to about 40 db. What is needed is more optimum point where the encoded stereo achieves at least 12 db separation between channels, since for the reason 25 explained above the listener may not be able to distinguish the difference even if channel separation was infinite, Given that a transformation or matrix coefficient in the encoder of 0.414 represents only 6db of stereo separation in the encoded media, it should be possible to reduce 30 this matrix coefficient to give 12db separation in the encoded signal. The present invention may provide a matrix encoder having improved separation between respective channels including between front and rear channels and between left and right channels.
6 SUMMARY OF THE INVENTION According to one aspect of the present invention there is provided an encoder for use in a surround sound system wherein at least four audio input signals (FL, FR, RL, RR) 5 representing an original sound field are encoded into two change signals (E, R) and said encoded two channel signals are decoded into at least four audio output signals (FL. FR' RU, RR) corresponding to said four audio input signals, said encoder including: matrix means connected to receive said four audio input signals for encoding said four input signals into two channel (L and R) output signals, said 10 matrix means including means responsive to said four input signals for producing L and R output signals as follows: L FL + kFR + JRL + jkRR R FR + kFL ~ jRR ~:kRL wherein k denotes a transformation or matrix coefficient having a value substantial 15 0.207 and j denotes a 90 degree phase shift. According to a further aspect of the present invention there is provided an encoder for use in a surround sound system wherein at least four audiAo input signals (FL, FR, RL, RR) representing an original sound field are encoded into two channel signals (L, R) 20 and said encoded two channel signal are decoded into at least four audio output signals (FL' FR' RU, RR') corresponding to said four audio input signals, said encoder including matrix means connected to receive said four audio input signals for encoding said four input signals into two channel (L and R) output signals. said matrix means including means responsive to said four input signals for producing L,<,, 25 and Rene output signals as follows: L FL + kFR + jRL + jKRR Reo FR + kFL -jRR - jkRL wherein k denotes a transformation or matrix coefficient having a value that is steered dynamically based on leve of rear signal (RL4RR) content relative to front signal 30 (FL+FR) content, Coefficient k may be steered from a first value to a second value. The coefficient k may be steered between the first and second values substantially linearly. The 7 coefficient k may have a first value that is substantially 0. The coefficient k may have a second value that is substantially 1A1 4. According to a still further aspect of the present invention there is provided an 5 encoding method for use in a surround sound system wherein at least four audio input signals (FL, FR, RL, RR) representing an original sound field are encoded into two channel signals (L, R) and said encoded two channel signals are decoded into at least four audio output signals (FLC, FR% RU, RR') corresponding to said four audio input signals, said method including: processing said four audio input signals into two 10 channel (L and R) output signals by matrix means responsive to said four input signals for producing L and R output signals as follows L = FL + kFR +jRL + jkRR R =FR + kFL -jRR kRL wherein k denotes a transformation or matrix coefficient having a value substantially 15 0.207 and j denotes a 90 degree phase shift According to a still further aspect of the present invention there is provided an encoding method for use in a surround sound system wherein at least four audio input signals (FL, FR, RL, RR) representing an original sound field are encoded into two 20 channel signals (L, R) and said encoded two channel signals are decoded into at least four audio output signals (FU, FR', RL, RR') corresponding to said four audio input signals, said method including: processing said four audio input signals into two channel (L and R} output signals by matrix means responsive to said four input signals for producing L, and R output signals as follows: 25 L. =FL + kFR + jRL + jkRR Re'v C FR + kFL -jRR -jkRL wherein k denotes a transformation or matrix coefficient and wherein said processing includes steering the value of coefficient k based on level of rear signal (RL+RR) content relative to front signal (FL+FR) content, 30 Coefficient k may be steered from a first value to a second value. The coefficient k may be steered between the first and second values substantially linearly The 8 coefficient k may have a first value that is substantially 0 The coefficient k may have a second value that is substantially 041 4_ The matrix means may include a plurality of components selected from adders, 5 multipliers, 90 phase shifters and comparators, DESCRIPTION OF A PREFERRED EMBODIMENT A preferred embodiment of the present invention will now be described with reference to the accompanying drawings wherein: 10 FIG, 1 is a block diagram showing principles of a "44-24" matrix system; FIG. 2 shows a configuration of a conventional encoder; 15 FIG. 3 shows a block diagram of a decoder including a variable matrix; FIG. 4 shows a graph of amplitude difference (d6) versus delay difference (mS) for illustrating the HAAS or precedence effect 20 FIG. 5 shows a configuration of an encoder according to an embodiment of the present invention; FIG. 6 shows a block diagram of logic associated with an encoder according to an embodiment of the present invention 25 FIG, 7 shows a block diagram of a multi-band encoder according to an embodiment of the present invention FIG. 8 shows a circuit diagram of a matrix encoder according to an embodiment of the 30 present invention; FIG. 9 shows a graphical representation of scaled values of k obtained from a scaling circuit; and 9 FIGS IA to IOD show examples of equal loudness response curves associated with a weighting filter Fig. 5 shows a matrix circuit 50 that is adapted to provide 12 d B separation between 5 decoded channels. Matrix circuit 50 includes a plurality of addersimultipliers and phase shifters arranged to produce encoded L and R output signals as follows: L FL + kFR + jRL +jkRR R= FR + kFL-jRR '-jkRL 10 wherein k denotes a transformation or matrix coefficient generally having a value approximately 0.207 and j denotes a 90 degree phase shift. The phase shifters may provide a substantially consistent phase shift over the entire audio frequency band. The four channel signals FL, FR. RU and RR' may be reproduced by a conventional 15 decoder as described in PCT application AU 201 0/001 666. it may be shown that when matrix coefficient k =0 20. separation between the encoded stereo L and R output signals is equal to at least 12 db, In addition, sepa rations between decoded channel FL' and adjacent channels FR and RL' are 20 respectively equal to 12 d B and separation between the channels FL and RR' in a diagonal direction equals infinity. This makes the system more balanced with no separation bias in the encoded and decoded signals, Testing performed wth the full decoder described in PCT/AU201 0/001 666 resulted in 25 12 db separation in the 4 surround output signals. Listeners could not hear a difference during the testing between the 12db matrix and the 40 db matrix or discrete surround sound, in addition Uisteners also could not hear a difference between the encoded surround stereo and normal stereo 30 Fig. 6 is a block diagram of a logic circuit for dynamically varying the matrix coefficient k. The logic circuit is suitable for steering a matrix encoder dynamically based on quantity of rear or surround signal content relative to front signal content. The dynamic logic circuit includes an equal loudness weighting filter 60 such as a modified Fletcher Munson/A-weighting or ITU-R 468 filter for providing compensation for I( variations in perceived loudness relative to frequency due to non linearity in human hearing response at least at some frequencies, The equal loudness weighting filter may be modiied to include a characteristic similar to a pink noise 1f) weighting at low frequencies, to further attenuate high amplitude low audibility sounds that may 5 otherwise unduly influence the steering logic circuit One reason for the compensation is that sounds in a 2-4 KHz octave appear loudest to the ear whilst sounds at other frequencies appear attenuated:weighting filters are sometimes used for the purpose of compensation, However, a pink noise filter is 10 preferred for music content over an A-weighting filter because the latter is mainly valid for pure tones and relatively quiet sounds. Pink noise is also known as 1/f noise, wherein power spectral density is inversely proportional to frequency. A pink noise contour gives greater attenuation at low 15 frequencies than a Fletcher Munson/A-weighting or ITU-R 468 weighting filter based on the tact that for equal power, amplitude is inversely proportional to frequency. Use of a pink noise contour may further reduce dominance of low frequency sounds (high amplitude but low audibility) in calculating steering logic values, which are based on amplitude, and results in better placement of sound information that may be important 20 for correct image generation. The dynamic logic circuit includes a mixer c for adding the compensated channel signals FL and FR to produce front sum signals (FL+FR) 62 and rear sum signals (RL+RR) 63, and comparator 64 for subtracting the two sum signals 62,63 to produce 25 a difference signal (FL+FR)-(RL+RR) 65. The difference signal 65 is applied to RMS detector 66. RMS detector 66 is adapted to compensate for the peak nature of music content. The averaging time constant over which RMS detector 66 measures a 'mean' value of a music signal preferably includes a first or 'attack time constant and a second or 'decay' time constant. The 'attack time constant may be substantially faster 30 than the 'decay' time constant in one example the attack time constant may be 20mS and the decay time constant rmay be 50mS for a full range RMS detector. in some embodiments an RMS detector including a single time constant may be used.
11 RMS detected output 67 is applied to logarithmic amplifier 68 to produce output 69 proportional to log|(FL+FRy(RL+RR)i. Logarithmic amplifier 68 is adapted to correct for logarithmic sensitivity of human hearing response to sound that spans a range of signal amplitudes or levels Output signal 69 is applied to scaling circuit 70 to produce 5 a scaled value 71 of transformation or matrix coefficient k based on a comparison of RMS detected and corrected signals 62 ant 63, In one form scaled value 7 1 may vary between 0.1 and 0414 representng a 20dB range between signals 62 and 63, Because it may be difficult to optimize scaled value 71 for all frequencies present in 10 music content, high and low frequency sounds may be scaled differently resulting in an unnatural reproduction of sounds for the listener, To mitigate against this the encoder of the present invention may include a multiband modification as shown in Figure 7, Fig. 7 shows a multi-band encoder wherein the audible spectrum may be split into 3 separate bands via band splitter 72. The bands include a low frequency 15 band A below 300 Hz, a midfrequency band B between 300-3KHz and a high frequency band C above 3KHz, Band splitter 72 may be interposed between input signals FL, FR RL RR and a variable matrix encoder comparable to encoder 15 (refer Fig. 1). A separate matrix encoder 73A, 73B, 73C may be used to produce a set of encoded output signals L,, and R for each frequency band A, B, C, The four 20 channel output signals for each band may be subsequently combined via band mixer 74, The output L may be obtained by combining contributions Lt. (A); Le. (B) L (C) produced by matrix encoders 73A, 73B and 73C respectively. The output Rpr' may be obtained by combining contributions R (A), Re () R (C) produced by matrix encoders 73A, 738 and 730 respectively. 25 When RMS detector 66 is used in a multiband decoder the attack time constant may be 3OmS and the decay time constant may be 60mS 'for band A . For band B the attack time constant may be 1OmS and the decay time constant may be 30mS. For band C the attack time constant may be 1 mS and the decay time constant may be 30 SS Fig, 8 shows a circuit diagram of a dynamic matrix encoder wherein the transformation or matrix coefficient k has a value that may be steered dynamically 12 depending on the level of surround or rear signal content (RL+RR) that is present relative to front signal content (FL +FR) The matrix encoder includes a dynamic logic circuit 80 for steering values of 5 coefficient k between OA and 0.41 4, and matrix circuits S and 82. Dynamic steering logic circuit 80 includes an equal loudness weighting filter 60 such as a modified Fletcher Munson filter: mixers 6 1a, ent, comparator 64 . RMS detector 66, logarithmic amplifier 68 and scaling circuit 70 as described above with reference to Fig. 6. Comparator 64 includes a differential circuit for producing a difference (FL+FR) 10 (RL+RR) signal 65 as described above RMS detector 66 has dual time constants as described above. Scaling circuit 70 may be implemented in software and/or hardware and may convert input logarithmic signal difference 69 to ramp values as illustrated in Fig. 9. 15 in Fig 9 the horizontal axis represents in dB, levels of surround or rear signal content (RL+RR) relative to front signal content (FL+FR). Thus the OdB point or level on the horizontal axis represents a balance between or equal front and rear signal content. Typically the X dB point or level on the horizontal axis may be substantially - 12dBi relative to front signal content but in some circumstances may be other than - 12d1 20 and may be determined based upon architecture of an implementation and/or discretion. in Fig. 9 the vertical axis represents scaled or dynamic values 71 of k. It may be seen that k has a first or minimum value of 0.1 when the relative signal content on the 25 horizontal axis is X dB or lower and has a second or maximum value of 0.414 when the relative signal content on the horizontal axis is OdB or greater. it may also be seen that the value of k increases substantially linearly from the first value 0.1 to the second value 0.414 as relative signal content on the horizontal axis increases from XdB to 0dB 30 Matrix circuit 1 includes summing amplifiers 83, 84, 85, multipliers 86, 87 and 90m phase shift circuit 88. The output L appearing at the output terminal of summing amplifier 85 and hence at the output of matrix circuit 8 1 is given by the following equation, 13 Lent = FL + kFR + j (RL + kRR) Matrix circuit 82 includes summing amplifiers 89, 90 difference amplifier 9 5 multipliers 92, 93 and 90Q phase shift Circuit 94, The output IR appearing at the output terminal of summing amplifier 9 1 and hence at the output of matrix circuit $2 is given by the following equation: R =FR + kFL -j (RR + kRL) 10 Equal loudness weighting filter 60 may include a modified Fletcher Munson -pink noise weighting filter including an ITU-R 468 weighting contour. Weighting filter 60 may be implemented in any suitable manner and by any suitable means. in one form the response of weighting filter 60 may include a frequency response contour as 15 shown in Fig IOD for a single band implementation. For multi-band implementations the response of weighting filter 60 may include frequency response contours as shown in Figs 1A to l0C for low band A, mid band B and high band C respectively. RMS detector 66 may be implemented in any suitable manner and by any suitable 20 means. In one form RMS detector 66 may be implemented on a digital sound processor such as a Texas Instruments TAS k 108 via Pure Path Studio Software. The invention described herein is susceptible to variations, modifications and/or additions other than those specifically described and it is to be understood that the 25 invention includes all such variations, modifications and/or additions which fail within the spirit and scope of the above description it may be appreciated that a matrix encoder as described herein may be applied to a surround sound system utilizing more than four audio input signals to represent an 30 original sound field, For example using the teachings of the present invention a pair of encoders as described herein may be applied to encode eight audio input signals rep resenting an original sound field into four channel signals and the encoded four channel signals may be decoded into eight audio output signals Such encoders may be applied to an installation including fou pairs of loudspeakers or speaker arrays 14 wherein each loudspeaker or speaker array :s arranged at a respective corner of a cube or a rectangular cuboid to define upper and lower planes of four loudspeakers or speaker arrays each, namely four loudspeakers or speaker arrays in the front and four loudspeakers or speaker arrays in the back. The upper plane of loudspeakers or 5 speaker arrays may be vertically separated relative to the lower plane of loudspeakers or speaker arrays by approximately 2-3m or other suitable distance depending on usable height in an associated listening zone or auditorium. The encoded four channel signals may be recorded on suitable media such as DVD, 10 BluRay disc or the like and/or broadcast via a HDTV transmission service such as Foxtel that :s capable of transmitting at least four channels of audio signals.