GB1586271A - 2-4 channel decoding matrix for regenerating quadrophonic signals - Google Patents

Description

(54) IMPROVED 2-4 CHANNEL DECODING MATRIX FOR REGENERATING QUADRAPHONIC SIGNALS (71) We, BRITISH BROADCASTING CORPORATION, a British Body Corporate of Broadcasting House, London, W1A 1AA, do hereby declare the invention, for which we pray that a patent may be granted to us, and the method by which it is to be performed, to be particularly described in and by the following statement:- This invention relates to improved 2--4 channel decoding matrices for regenerating quadraphonic signals from the output of a 2-channel transmission or recording system.

Our British Patent Application No. 34839/74, now Patent No. 1,526,195 (our earlier application) describes and inter alia claims a channel reduction matrix for generating from input audio signals a 2-channel representation capable of quadraphonic, stereophonic or monophonic audio reproduction, wherein the output signals in the two channels either actually comprise or can be notionally represented by two matrixed signals which may vary in amplitude and relative phase, each output signal being formed of combinations of the input signals with such weighted amplitudes and phases that when the input signals represent a sound source moving in a circle around a datum position the arc tangent (a/2) of the ratio of amplitudes and the relative phase (A) of the two said matrixed signals for input signals representing a sound source at the centre front, right and left front, centre right and left, right and left back and centre back positions are within the ranges described with reference to the Table therein and illustrated in Figure 10 of the drawings accompanying the Complete Specification of that application.

The preferred encoding matrix of our earlier application is known as matrix Hs For use in decoding our earlier application describes with reference to Figure 19 of the drawings accompanying the complete specification thereof a decoder which uses a non-linear decoding matrix of the type described in British Patent No. 1,402,320 in conjunction with a 60 degree phase shift circuit connected before the input terminals.

Although the quadraphonic performance of that decoder is good, it still has limitations.

This invention is concerned with improved non-linear or logic enhanced decoding matrices for decoding a 2-channel signal which has been encoded by means of matrix H or a matrix substantially similar to it, and which make use of the advantageous properties of matrix H.

The invention, which is defined in the appended claims, will be described, by way of example, with reference to the accompanying drawings, in which: Figure 1 is a block circuit diagram of a first decoder embodying the invention; Figure 2 shows the variation law of the logic signals of the decoder; Figure 3 illustrates typical separation curves for one source; Figure 4 is a block circuit diagram of a second decoder embodying the invention; and Figure 5 illustrates the performance of the decoder of Figure 4.

In this specification the following symbols are used with the meanings indicated: LF,RF,LB,RB quadraphonic origination signals LTFS RRFS LTBS R B decoded quadraphonic signals Lr RT matrix encoded 2-channel (stereo) signals

f front b back I left # logic control signals right r# polar representation of vector, of modulus r and argument 0 degrees The matrix H decoding equations are described in our earlier application and can be written using the above notation as follows:: L'F=0t940 LT /-20C+0.342 RT 55 R'F=0.342 L,/-55 +0.940 RT /20 L'B=0.940 LT 25 +0.342 RT /- 115 R'B=0.342 LT/1150+0.940 RT /25 (1) These decoding equations can be rewritten as: L'F=[0.940)f(LT-RT /750)+1 1.282 RT S] /-200 R'F=[-0.940 -f (LfR7RT /750)+ r 1.282 LT] -o L'B=[0.940 b (LT+RT j400)-I 1.282 RT 40 ] 25 R'B=[0.940 b (LT+RT /40 )-r 1.282 LT]/65 (2) where, for linear decoding, the logic signalsf b, 1, r are at their quiescent value of unity. By varying the logic signals in a similar way to that described in Patent No.

1,402,320, the decoder of which has been commercially produced by the patentee under the trade mark "Variomatrix", high separation for a principal source can be achieved.

It will be appreciated that other values of absolute amplitude and phase may be taken provided that the inter channel amplitude and phase differences are maintained.

In accordance with this invention we propose two advantageous decoding circuits making use of the matrix H decoding equations. Figure 1 is a block diagram of a first such embodiment of the invention. This has two inputs 10 for receiving the 2-channel input signals LT, RT encoded in accordance with our earlier application, and four outputs 12 for supplying the 4-channel output signals LF, R'F, Llg and R',,.

The upper part of the circuit as illustrated implements the matrix represented by the four equations (2) above, and the lower part generates the logic signals. The logic signal generation will be described first.

The input signals LT and RT are first applied to respective accurate wide-band all-pass phase shift networks LS R which provide a relative phase shift between the signals Lr and RT. Circuit Q is simply a compensating phase shift of 0 , but circuit R provides four outputs shifted by 220, 40 , 67 , and 75 respectively. Of these the 22 and 67 signals are used in the logic circuitry and the 40 and 75 signals in the matrix circuitry. The three signals LT /0 , RT/22 and RT /670 are applied to respective high pass filters (h.p.f.).A phase detector receives the signals LT /0 and Rr /670 and compares the phase of these two signals to provide two balanced d.c. outputs which vary inversely with each other. The phase detector consists of two high-gain limiting amplifiers and a phase discriminator. The outputs are proportional to the phase difference, with 900 representing a quiescent value. These signals are applied to filters which determine the response times, i.e. the attack and decay times, of the signals to changes in the relative phase of the input signals LT, RT. Finally a "law" circuit transforms the signals proportional to phase to the logic signalsf and b, which conform approximately with the following equations.

If the argument (phase difference) of LT/RT is 0, and #-67 is defined as A in degrees, then: (i) for Ac90 f=1.7-0.7 A/90 b=0.7+0.3 A/90 (ii) for A > 900 f=1.3-0.3 A/90 b=0.3+0.7 A/90 Thus the logic signals f and b are each linear with A over two segments as shown in Figure 2(a), and vary between about 0.7 and 1.7.

The signals LT /0 and RT /220 are applied to a level detector, which may consist of phase shift circuits, sum and difference circuits, and a phase discriminator as shown in Figure 13 of Patent 1,402,320. This likewise produces two balanced signals which after filtering are applied to a "law" circuit. This law circuit is shown in Figure 1 as having four outputs lf, r Ib, and rb, but for present purposes it may be assumed that 11,=1 and rt=rb=r.The logic signals I and r bear approximately the following relation to the input signals LT and RT. If tan-' LT/RTI is defined as T degrees, and T varies between 0 and 90 , then: (i) for LT < RT i.e. T#45 1=0.7+0.6 T/90 r=1.7-1.4 T/90 (ii) for LT2RT i.e. T > 450 1=0.3+1.4 T/90 r=1.7-1.4 T/90 Thus the logic signals I and r are each linear with T over two segments as shown in Figure 2(b), and vary between about 0.7 and 1.7. Extreme values of up to 0.6 and 1.8 could be used, however.

It may be that further variations can be made to the "law" equations where particular changes to the signal separations would be advantageous. In particular, the extreme values of the logic signals b, I and r may be altered slightly so as to be closer to unity. Typical alternative values might be 0.8 and 1.4. With these alterations, the programme dependence of the decoding is reduced, and the decoding is more linear.

The low pass filters preceding the law circuits operate such that there is a relatively quick response to a new input signal but a less rapid response when the signal is removed. For example an attack time of about 10 milliseconds and a decay time of some tens of milliseconds may be appropriate. Thus it will be appreciated that the above equations for the logic signals represent the steadystate situation.

The matrix portion of the circuit will now be described. Six adder/scalers 14 are arranged as- illustrated to generate from the signals LT 0 (=LT) RT /40 and RT 75 the following signals: 0.940 (LT-RT [75 1.282 RT [75 1.282 Lr 0.940 (LT+RT /400 1.282 RT/40 1.282 LT.

The third and sixth circuits may be combined. These six signals are then applied to voltage-controlled amplifiers (v.c.a.'s) which receive respectively the logic signals f 4, rf, b, 4, and r,. Four summation circuits 18 combine these six signals, as illustrated, to produce, respectively:: 0.940f(LT-RTiu50)+1.282 If RT /75 1.282 rf LT- 0.940 f (LT-RT-75 0.940 b (LT+RT /40 )-1.4282 l0 RT 40" 0.940 b (LT+RT /40 )-1.282 rb LT Finally four output phase shift circuits L'F, R'F, 8L', and 8R', provide relative phase shifts of -20 , -55 , 25 and -65 to generate the output signals LUFFS R' F' L'B and R',, respectively.

The voltage-controlled amplifiers are frequency-dependent. At low audio frequencies they have constant gain and there is no logic enhancement so that the decoding matrix is simply the basic linear decoding matrix H. The cross-over frequency may be about 200 Hz so that audio signals below the maximum frequency of the logic signals, determined by their attack time, are not subject to logic control.

The separation of the logic signals I and r into 4 and 4, and rf and rb permits subjective empirical adjustment of these signals. In all cases the quiescent values of the logic signals are unity, which value they also take in the absence of any input signal (i.e. Lr=R=0).

Analysis of the variable matrix can be performed and maximum separation figures have been predicted. This has shown that adequate separation can be achieved for most source locations, but for a corner signal the maximum front-toback separation is relatively low (13.6 dB) and this may displace the image slightly.

Separation can be increased by slightly altering the front and back phase-angles of the RT signal from their values of 75" and 40 respectively. With this modification, a better overall performance can be expected.

In the decoder of our earlier application, the matrix is independent of logic action at low frequencies and reverts to the basic matrix. However, at these low frequencies, the basic matrix can be distorted by left-to-right blend-circuits in the output channels of the "Variomatrix" which localise low-frequency sounds on the front/back centre-line. No such blend circuits are included in the Matrix H decoder illustrated in Figure 1, since basic Matrix H is capable of accurately locating sounds without logic enhancement.

This has the added advantage of maintaining high separation to a lower frequency and, as a result, the total energy of the crosstalk signals is less for the same maximum separation figure (at 1 kHz). This permits a reduction of the logic action so as to reduce image wandering, whilst still maintaining adequate separation.

The frequency response is shown in Figure 3 in which the solid lines show the response of the decoder of Figure 1 for an LF signal source and the dashed lines show the departure from this of the decoder of our earlier application.

The logic signals modulate the audio signals and thus the removal of the blend-circuits might be expected to result in audible intermodulation distortion.

However, in practice no evidence of this has been found, possibly because the distortion is of a transient nature.

At high frequencies (above 1 kHz), the high separation of mid-band frequencies is maintained by not restricting the high-frequency response of the voltage controlled amplifiers in the variable-matrix.

Work on the effects of interchannel phase-differences on the localisation of quadraphonic images has shown that even small phase-differences (of the order of 20 ) can sometimes displace or blur an image. There is also evidence that adverse phase-differences can increase the audibility of image wandering in the following way. If a large phase-difference exists between two principal-source signals (for a CF source, say), the image is displaced and even small additional variations of phase can cause the image to wander. If, on the other hand, the phase-difference is small, the same variations will have a negligible effect.

Care has, therefore, to be taken in the design of the output phase-shifter circuits, to ensure that the proper phase relationships exist between two principalsource signals, and between a principal-source signal and a crosstalk signal. The output phase-shifts F' F' LB 8,, and 8,, in one decoder were slightly changed from the values shown in equations (2) above to --200, -60", +30", -75" respectively in order to account for the logic action and the higher interchannel separation produced. They are accurate up to a frequency of about 4 kHz, it being unnecessary to maintain stringent tolerances at higher frequencies (unlike the input phase-shifters in the decoder).With this decoder, images were found to be sharper and better defined than those given by the modified "Variomatrix" of our earlier application; further, slightly less image wandering was also noted, although the sharper images might have been expected to emphasise this effect.

On a brief subjective assessment of this decoder, significant improvements were found as compared with the modified "Variomatrix" of our earlier application. The principal improvements consisted of sharper images, a greater sense of 'openness' and better overall perspective, fewer sibilant mislocations, and a greater tolerance to listener position.

It is thought that this type of decoding, which uses different interchannel phase-angles (400 and 75 ) for decoding the front and back channels, could be optimised by slightly altering the phase angles from those used in the basic Matrix H decoding equations. At all times, however, both phase angles should lie within the limits of 40 and 80". Likewise, other parameters of the decoder can be altered.

The relative gains of the adder/scalers 14 may vary by up to +20 /". The phase lead of the Rut /670 signal to the logic circuitry may vary in the range 500 and 80" relative to the LT /0 signal, whilst the phase of the RT=. signal differs from that of the RT /67 signal by 45" (a200). The optimum values of the output phase shifts depend on the particular input phase angles and gains used, but they would normally lie within f40" of the following values:: LF=200 LB=25 #Rf=-500 8RB=-95 Further developments can also be envisaged, for example, using more complex forms of logic-enhancement, or the incorporation of delay-lines in the audio channels to overcome low-frequency localisation and transient problems.

The Matrix H decoder described above is a fairly complex device as Figure 1 shows. Figure 4 shows a simplified decoder with a performance comparable with, if not better than that of the Matrix H logic decoder described above. Much of Figure 4 is similar to Figure 1 and only the differences will be described in detail.

The logic circuits used are the same as those incorporated in the Matrix H logic decoder of Figure 1, but the basic matrix is modified to decode using a single phase-angle (67 ) for both front and back channels. The decode equations become:- L F=[0 940 f (LTRT )+ l 1.282 RT /670] R'F=[-0.940f(LT-RT l2)+r 1.282 LT]/50 L'B=[0.940 b (LT+RT /67 )-l 1.282 RT /670]/250 R',,=[0.940 b (LT+RT /67 )-r 1.282 LT]l95 (3) where the basic matrix is given when f, b, 1, r=l.

Since the same phase shift of 67 is applied to RT for both the front and back channels, the matrix portion of the circuitry is much reduced in complexity. In view of the previous detailed description of Figure 1 the construction of the circuitry will be readily apparent from Figure 4. It should be noted however that the output phase shifts LF7 RF, LB and #Rb are slightly changed to --20 , --50 , 250 and -95 respectively.

We have found that with this circuit separations for corner sources are improved, without significantly sacrificing other locations. Otherwise the performance was subjectively similar to that of the decoder of Figure 1. The output phase-shifts used give good phase corrections both with and without logic enhancement. Typical performance figures are shown in Figure 5, in which the response to the eight cardinal positions is illustrated. Each small square describes the decoded output signal relationships to the appropriate cardinal positions; thus the top-centre square refers to the decoded output signals for a CF (centre-front) encoded input signal.

It will be appreciated that there is some latitude in the exact values used for the phase-shifts and amplification factors. Typically, for the single phase-plane decoder of Figure 4, the RT input phase angle of 67 may vary between 75 and 500, whilst at the same time varying the RT/220 phase to differ from the RT /67 phase by 45 (#20 ). The other tolerances are the same as those of the double phase-plane decoder of Figure 1 given above. The best performance is likely to be given with values close to those specified, and any final solution should be a compromise as described earlier.

Thus we have provided decoders based on the advantageous matrix H of our earlier application which exploit the possibilities of logic enhancement to provide quadraphonic reproduction of subjectively good quality. In particular, when used with matrix H of our earlier application, it is possible to maintain high separation (e.g. 15 dB) between independent source signals over a wide frequency range of typically 400 Hz to 8 KHz.

WHAT WE CLAIM IS:- 1. A 2--4 channel decoding matrix for regenerating quadraphonic signals from the output of a 2-channel transmission or recording system, comprising two input terminals, four output terminals, and signal processing means connected between the input and output terminals, the processing means being adapted to operate on input signals LT and RT received at the input terminals to generate signals UF, R'F, L', and R',, at the output terminals, wherein, over at least a substantial portion of the frequency range of the input signals:

**WARNING** end of DESC field may overlap start of CLMS **.

Claims (5)

**WARNING** start of CLMS field may overlap end of DESC **. LF=200 LB=25 #Rf=-500 8RB=-95 Further developments can also be envisaged, for example, using more complex forms of logic-enhancement, or the incorporation of delay-lines in the audio channels to overcome low-frequency localisation and transient problems. The Matrix H decoder described above is a fairly complex device as Figure 1 shows. Figure 4 shows a simplified decoder with a performance comparable with, if not better than that of the Matrix H logic decoder described above. Much of Figure 4 is similar to Figure 1 and only the differences will be described in detail. The logic circuits used are the same as those incorporated in the Matrix H logic decoder of Figure 1, but the basic matrix is modified to decode using a single phase-angle (67 ) for both front and back channels. The decode equations become:- L F=[0 940 f (LTRT )+ l 1.282 RT /670] R'F=[-0.940f(LT-RT l2)+r 1.282 LT]/50 L'B=[0.940 b (LT+RT /67 )-l 1.282 RT /670]/250 R',,=[0.940 b (LT+RT /67 )-r 1.282 LT]l95 (3) where the basic matrix is given when f, b, 1, r=l. Since the same phase shift of 67 is applied to RT for both the front and back channels, the matrix portion of the circuitry is much reduced in complexity. In view of the previous detailed description of Figure 1 the construction of the circuitry will be readily apparent from Figure 4. It should be noted however that the output phase shifts LF7 RF, LB and #Rb are slightly changed to --20 , --50 , 250 and -95 respectively. We have found that with this circuit separations for corner sources are improved, without significantly sacrificing other locations. Otherwise the performance was subjectively similar to that of the decoder of Figure 1. The output phase-shifts used give good phase corrections both with and without logic enhancement. Typical performance figures are shown in Figure 5, in which the response to the eight cardinal positions is illustrated. Each small square describes the decoded output signal relationships to the appropriate cardinal positions; thus the top-centre square refers to the decoded output signals for a CF (centre-front) encoded input signal. It will be appreciated that there is some latitude in the exact values used for the phase-shifts and amplification factors. Typically, for the single phase-plane decoder of Figure 4, the RT input phase angle of 67 may vary between 75 and 500, whilst at the same time varying the RT/220 phase to differ from the RT /67 phase by 45 (#20 ). The other tolerances are the same as those of the double phase-plane decoder of Figure 1 given above. The best performance is likely to be given with values close to those specified, and any final solution should be a compromise as described earlier. Thus we have provided decoders based on the advantageous matrix H of our earlier application which exploit the possibilities of logic enhancement to provide quadraphonic reproduction of subjectively good quality. In particular, when used with matrix H of our earlier application, it is possible to maintain high separation (e.g. 15 dB) between independent source signals over a wide frequency range of typically 400 Hz to 8 KHz. WHAT WE CLAIM IS:- 1. A 2--4 channel decoding matrix for regenerating quadraphonic signals from the output of a 2-channel transmission or recording system, comprising two input terminals, four output terminals, and signal processing means connected between the input and output terminals, the processing means being adapted to operate on input signals LT and RT received at the input terminals to generate signals UF, R'F, L', and R',, at the output terminals, wherein, over at least a substantial portion of the frequency range of the input signals: and the following seven conditions are satisfied: (i) 0 and Oz each lie within the range from 400 to 800, (ii) K2/K,=(1.282/0.940)# up to 30%, (iii) f, b, 1,, 1b' r, and rb each vary between an upper extreme value which is not greater than 1.8 and a lower extreme value which is not less than 0.6. (iv) f and b are both functions of the phase difference between LT and RT /&alpha;, where a lies in the range 400 to 800. (v) lf, lb, r, and rb are all functions of the ratio of the amplitudes of LT and R (vi) The relative values of #lF, #RF L and ,, are respectively -20 , -50 , 25 and -95 , each plus or minus up to 400. (vii) The matrix is such as to generate quadraphonic output signals from a signal encoded in accordance with Claim 1 of our earlier Specification No.

1,526,195 (as published).

2. A matrix according to Claim 1, wherein the values of QF, #RF, #LB and ,F are substantially -20 , -50 , 25 and -95 respectively.

3. A matrix according to Claim 1, substantially in accordance with equations (3) herein.

4. A 24 channel decoding matrix substantially as herein described with reference to Figure 1 of the accompanying drawings.

5. A 24 channel decoding matrix substantially as herein described with reference to Figure 4 of the accompanying drawings.