US3183310A - Bandwidth compression and expansion system for transmission of speech - Google Patents

Description

May 11, 1965 Filed June 5, 1962 D. GABOR 3, BANDWIDTH COMPRESSION AND EXPANSION SYSTEM FOR TRANSMISSION OF SPEECH 2 Sheets-Sheet 1 INPUT TR DISCRIM.

&SELECTOR 9 M TRANSMIT. 5

I 52 5; 555 E 55 i 3 f 1111121: :Ii:

JNVE/VTOR DENN l S GABOR B h/fnman Wm M ATTO PNEKS' May 11, 1965 D. GABOR 3, BANDWIDTH CQMPRESS AND EXPANSION SYSTEM Filed June 5,- 1962 OR TRANSM ION 0F SPEECH v 2 Sheets-Sheet 2 FlG06 0 INVENTOR DE NNIS GABOR A T TORNEYS United States Patent 3,183,310 BANDWIDTH COMPRESSION AND EXPANSION SYSTEM FOR TRANSMISSIGN 0F SPEECH Dennis Gabor, London, England, assignor to National Re if search Development Corporation, London, England, a

British corporation I Filed June 5, 1962, Ser. No. 200,243 7 Claims. (Cl. 179-1002) This invention relates to the transmission and/or recording of speech and is concerned with methods of and apparatus for reducing the frequency band required therefor.

In British Patent No. 616,156, there is described a frequency compression systemvin which a record such as a sound film or a magnetic tape record is taken off the signal and this is processed by scanning it by means of pick-ups, running with the record, but at' a somewhat smaller speed. The difference between the speed of the record and the scanning speed produces a reduction of the frequencies by the Doppler effect, and the compressed signal occupies a reduced waveband in the transmission. In the receiving apparatus a new record is taken and this is scanned by pick-ups running at a speed higher than that of the record so as to restore the original frequencies. The full theory of this process is given in D. Gabor, Theory of Communication, vol. 93, part III, Journ. I.E.E., pp. 445457, 1946. i

As explained in New Possibilities in Speech Transmission by D. Gabor, l.I.E.E., vol. 94, part III, p. 369,

1947, simple frequency compression as described in the above-cited British patent, suffers from various distortions. The reproduced speech sounds eitherrnonotonous or rough, and at high compression ratios the intelligibility may suffer from short consonants dropping out of the speech. Inthe second quoted publication an improvement on the simple method is described, called frequency compression with pitch-controlled scalej Realisations of this principle are described in British Patents Nos. 645,542 and 682,323.. v

As explained in the above publication, the processing of the signal to reduce its bandwidth is carried out by repetitive scanning of a record by a plurality of pick-up devices as the record is progressed past a window, i.e., a region in which the record is accessible to the pick-up devices, the signals from each pick-up device being varied during the scan from Zero through a maximum back to zero in accordance with a predetermined law and being combined I additively for application to the transmission or recording channel. The reciprocal of the time between the scanning of any point on the record by two successive piclc up devices will be called the .recurrence frequency. Now any particular frequency in the signal gives rise in the combined output from the scanning pick-up devices to a spectrum of frequencies, the form of which is dependent inter alia upon the recurrence frequency. For some fre-' quencies the frequency imposed by the scanning technique itself appears in the spectrum of'outputfrequenciejs and such a frequency is described asa preferredfrequency. Such preferred frequencies in other words represent those frequencies which, if present in. the pick-up signal give rise to spectra in which the preferred frequency itself.

is present and it is present at the, greatest amplitude.

Because of this effect it wasfound that bandwidth com-i pression systems according to the earlier of the prior-pro:

has been referred to above as frequency compression with pitch-controlled scale This means that in these systems thepitch ofthe voice is extracted from the signal and used to control the values'of the preferred frequencies so that 3,183,310 Patented May 1 1,1965

the pitch frequency is always represented in the spectra of output frequencies and may be a preferred frequency.

This can be done by suitable control of the recurrence.

frequency as described in British Patent No. 645,542 for example. In the prior proposals variation of the preferred frequencies in a continuous manner was proposed.-.

According to this invention apparatus'for the processing of speech signals to produce compressionot expansion of the bandwidth thereof comprises means for producing a plurality of separate records of the speech signals to be. processed, a plurality of sets of pick-up devices, eachset comprising a plurality of pick-up devices and each set being associated with a different record, said sets operating to scan the associated records at different eifectiverecurrence frequencies, means for combining the outputs of the picloup devices of a set to produce a processed signal consisting of a reduced or expanded bandwidth version of the speech signals, and discriminating means for continuously selecting one of said processed signals'as an output signal in accordance with the pitch component of said speech signals. 1 A

The various sets of scanning pick-up devices may consist of different numbers of individual pick-up devices) whereby each set provides a different effective recurrence frequency when scanned past therecord at a given speed.

The record may be a temporary record, for example on magnetic tape and may take the form of a multiple record each version of which is scanned by a different set of pick-up devices so that the output may be taken fromthe appropriate set in signal.

According to a feature of the invention the scanning accordance with the pitch of the of the record may be effected in such a way that each, successive pick-up device becomes operat'ively engaged FIGURE 1 is 'a schematic diagram of a complete trans mission system embodying the invention;

FIGURES 2, 3 and 4 are three views of one device according to the invention, preferably used as a frequency compressor;

FIGURE 5 is a schematic developed view of the periphery of the conical drum of the device shown in FIGURES 2, 3 and 4 showing the scanning gaps of the magnetic heads;

FIGURE 6 explains the gradual engagement anddisengagement of the magnetic pick-ups with the tracks on the magnetic tape;

FIGURES 7 and 8 are two Window-curves, that is to say they show the engagement of a pick-up disk with its corresponding record track as a function of time; and

FIGURES 9 and 10 are sectional views of a variant of the invention in which the conical drum is hollow and scans the magnetic tape from the outside.

FIGURE 1 is a schematic block diagram of the whole of a transmission system embodying the invention, consisting of transmitter and receiver. The speech signal to be transmitted is fed through an input amplifier 1 to magnetic re cording heads, which are connected in parallel and ar-- ranged in one line across a magnetic tape 2. At the same time the speech signal is fed into a pitch discriminator. 3,.

well known in the art, which separates. the pitch from the rest of thesignal, and measures'its value continuously,

This unit contains also a selector, of a type well known in the electronic art, which opens one or other of a set of relays 4 which control the input to the transmitter amplifier 5. The magnetic tape .2 passes partly around a conical drum 6 rotated to produce slip between the tape and the surface of the drum and is there associated with scanning heads in a manner to be described later which provide bandwidth-compressed versions of the speech signal at the slip rings 9. The outputs derived from slip rings 9 are applied to relays 4 so that the appropriate one is selected for application to the transmitter amplifier 5. The output selected is that whose repetition frequency corresponds most-closely to the instantaneous pitch of the speaking voice, according to the rules to be explained below. In the case of unvoiced sounds the relay which was last operated remains operated until the pitch again becomes ascertainable. When the pitch changes, so that the signal selection has to be transferred from one slip ring to the other, a convenient overlap is arranged between the relays, of the order of a few milliseconds, in such a way that there is no discontinuous break in the transmitted intensity. Fast mechanical relays can be used for this purpose, but it is understood that any electronic equivalent such as vacuum valves, gas filled valves or transistors may serve instead.

The receiver consists essentially of the same component parts indicated by the same reference numerals sufiixed with a 1. In the receiver, however, the bandwidth compressor 6 is of course replaced by an expander 6. It may be, however, unnecessary to use a pitch discriminator, if the pitch as measured in the transmitter is directly transmitted, as an amplitude or in some coded form, to the receiver. In this case only a selector is required. It is known that a frequency band of 25 cycles/sec. is sufiicient for the continuous transmission of the larynx frequency of a human voice.

FIGURES 2, 3 and 4 show in more detail the bandwidth compressor 6 of FIGURE 1, wherein 2 is the magnetic tape, 12 is a multiple magnetic recording head, of conventional design, but with spacings between the tracks at least equal to the width of one track. 13 is a rotating drum, composed of disks 14 of which one is shown in section in FIGURE 2. Each disk contains a number of magnetic heads 15, with windings 16, which are connected in series by a wire housed in a circumferential groove 17. The ends of the windings of each ring are led through one of the longitudinal bores '18 to slip rings 9, on which run brushes 10. The tape is driven by a capstan 21, in conjunction with a resilient roller 22. The tape is joined up to form a closed loop and runs through an erasing head not shown in the figure. The device, as shown in FIGURES 2, 3 and 4, is particularly suitable for use at the transmitter, because in the compressor the tape runs in the same direction as the drum but with a small difference in the velocities, and for reasons to be explained later, the engagement angle is preferably large so that many heads engage simultaneously with the tape.

It is seen in FIGURE 2 that the magnetic tape is recorded from the side opposite to that from which it is scanned. This makes it possible to put the recording head 12 very close to the point at which the scanning by the rotating heads starts and recording or scanning from the back has no disadvantages, because as will be shown later in a numerical example, the shortest wavelengths to be recorded or scanned are so long in this device that the tape can be almost equally well recorded or scanned from either side. One can operate the tape with the oxide coating at the outside or in contact with the drum.

FIGURE is a schematic developed view of the surface of the drum 13 showing the gaps of the scanning heads. There are eight disks in the present example and the numbers of the heads in these are 8, 9, 15. The total number is 92. This is considered as about the smallest number for satisfactory operation. As an example, assume that the speed of rotation of the drum is 750/rninute or 12.5/sec. The repetition frequencies, that is to say the 4 number of passages of gaps before a point fixed relative to the various disks will then be 100, 112.5, 187.5/sec.

FIGURE 6 is a view of the tape showing that part of the drum surface contained between the two lines C --C on which the tape makes contact with the drum. 'The shaded stripes 23 are the recorded tracks and the oppositely shaded curved stripes 2 4 are the paths of the scanning heads over the tape. R is the position of the recorder 12. It is seen that first the disks are out of contact with the tracks and engage gradually to a maximum, to disengage again.

The engagement or window curve corresponding to this arrangement is shown in FIGURE 7. Assuming that a constant amplitude is recorded on the tape, the amplitude A picked up by a head is the elliptic curve as shown.

FIGURE 8 shows a window-curve corresponding to the case of a gap of a scanning head wider than the track width and the relative transverse motion of gap and track also larger than the track Width. This produces a flattopped Window-curve, which has an advantage particularly in the receiver in which only a few gaps are simultaneously under the window, as will be explained later.

FIGURES 9 and 10 are a face view partly in section and a cross-section of a device in which the conical drum is hollow and the magnetic heads scan the tape from the outside. This modification has a particular advantage as a receiver, that is to say bandwidth expander, because in the expander the tape and the scanning heads move in opposite directions, and with the tape at the outside this might lead to instabilities in the motion of the tape. The external cone as shown in FIGURE 9 is also particularly advantageous in the receiver, because in this, as will be shown later, the angle of scanning is smaller than. in the transmitter. This modification has the further advantage that it is possible to operate it with the tape out of contact with the rotating head.

In FIGURES 9 and 10 the conical drum and its constituents correspond to those in FIGURES 2, 3 and 4, but the whole drive of the tape is arranged inside the conical cavity. In addition to the recording head 12 and the drive consisting of the capstan 21 and the roller 22 the drawing shows also an erasing head 25. Between 12 and 22 the tape runs over the conical saddle-piece 26, which is preferably highly polished, and is arranged at a close distance from the conical drum surface, so that the gap between the tape and the drum is of the order of a few thousandths of an inch, which can be achieved without difliculty. Between the capstan and the eraser the tape is fed into a magazine 27. This has the advantage that a greater length of tape can be accommodated in the loop, which reduces wear and also minimizes the effect of the joint.

Having now explained the elements of the invention, the operation may be further elucidated with the aid of a numerical example, in which it will be assumed that the compression ratio is n=%. It has been shown in detail in the above quoted publications, but is also immediately evident, that in order to achieve this in the transmitter, in which the tape speed is v, the linear speed of the drum must be 14 A1 v so that the relative speed between the scanning head and the tape is v 8, while in the receiver in which the tape speed is v the scanning speed must be u =7v the negative sign indicating that the drum and the tape run in opposite directions.

It might be thought that this condition cannot be realized with conical drums, because the speed of the tape is the same in all tracks, while the peripheral speed of the scanning heads is proportional to the radius of the drum. The difiiculty is overcome according to the invention as follows:

Make the speed ratios correct for the central ring of heads in the compressor and in the expander, that is to say let u /v u /v =-7 for the said central rings. In

that case the head-rings at the extreme outside and inside will have speeds u (1:n in the compressor and in the expander, where 1 and 1 are the differences between the largest and the mean radii, divided by the mean radii, in the compressor and in the expander respectively. The resulting frequency conversion ratio will then have instead of unity the value By making or, more generally, by making the ratio of the errors in speed in the expander and in the compressor equal to the expansion ratio the second term in the above expression becomes zero, and only the last term remains, which, as will be shown, can be made very small. I

Let W and W be the widths of the tapes in the transmitter and in the receiver, r r the mean radii, and 1 41 the semi-angles of the cones. We have then Let W1, W2 be the track widths in the transmitter and in the receiver, and 20 the angles on which the tape is in contact with the periphery of the drums. We have then the condition that on the arcs r 9 and r 6 the tape must move transversely by W1 and W2 respectively relative to the scanning ring, so as to be disengaged at the start, and fully engaged in the centre of the scanning window. This givesthe conditions This is to say that in order to obtain minimum frequency distortion on all tracks the arc of contact in the expander must be A :0.3155 times that in the compressor.

It will now be shown in an example that this in fact results in negligibly small errors in the final reproduction of frequencies. Let 0 =f /zm as in FIG. 1, r =r 50 mm., hence 0 =0.355.' /21r=O.56 or 32, corresponding to a total contact angle of 64 in the expander. Let w =w =1 mm., W W =20 mm. By the condition H this gives sin =0.0162 (about 55) and sin =0.112-9 (about 725). The error co-eflicients now become q =0.00324 and n =0.025 8. The maximum error in the reproduction of frequencies is therefore 8 which is negligible. Hence, in spite of the conicity of the drums the resulting reproduction'of frequencies can be made very good.

X 0.0 0324 0.0258 5.1 X l0 or 0.05%

The further rules for the dimensioning of the compres- 6. The averagepitch of male voices is cycles/sec., and the lowest pitch P may be assumed as 114.28 cycles/sec, which is a deep bass voice. If this is to be transmitted without distortion, the recurrence frequency in the transmitter must be 100 cycles/see, or a sub-multiple of it such as 50, 33.3, 25. The best choice is 100 cycles/see, because, other things being equal, this gives the largest number of heads simultaneously scanning .the tape in the compressor, and the larger .this number, the better the ratio of scanned to' unscannedfraction of the tape. We assume therefore a recurrence frequency R=100 cycles/sec. both inthe compressor and in the receiver. This gives undistorted transmission of P because in the compressor we thereby produce a beat frequency which is Aaof P =114.28, while in the expander we perform the operation which restores the original frequency. Moreover, as shown in my above quoted publications, this frequency will be produced almost pure, provided that the ntunber of heads simultaneously scanning the tapeis sufficiently large. it these numbers are too small, combination tones of the desired frequencies with multiples of 100 cycles/sec. will also appear, and distort thetransrnission.

114.28, 128.4, 142.8, 157.0, 171.1, 185.5, 200.0, 214.0, 228.56,. sec. 1

which are in the ratio of 8:9:10 15.; i

This scale as good as completely covers the range of male larynx frequencies, while the average female pitch is about 230. At a pitch of twice the lowest male pitch the selector therefore goes back to the first ring, and the correct compression is produced by the beat frequency which is /s of 228.56. As shown in my previously quoted publications, this beat frequency is produced automatically and in almost pure form if the selector selects the .first ring with R=100 cycles/ sec. and the same. applies to even higher female larynx frequencies.

. The repetition frequency R=l00/ sec. with the number N=8 determines the turning speed of the compressor drum, which must be 12.5 revs/sec. or 750 revs/min. With the previously assumed radius of r =50 mm. of 2" this gives a peripheral speed of-the compressor drum of 157"/sec., and a tape speed of times 'asmuch, i.e., /sec. At such a large speed the minimum wavelength which is of importance in speech. transmission,

which is about 4000 cycles/ sec. will be (1045f which fully justifies the recording or scanning of the tape from the back, as shown in FIGURES- 2, 3 and 4 and ,9 and 10.

. With 8 heads and the assumed contact. angle of 180 there are 4 heads simultaneously scanning the tape, and as they'move relatively to the tape-with' flz of its speed, they will scan exactly 50% of it. Experience shows that this is fully sufiicient. Thelength of thenot scanned portions of the tape is A sec. or 5 millisecgwhich is too short for even the shortest consonant to fall into the gap; Moreover this time shortens to 2.67 msec. at the thicker end of the drum where the number of heads is 15.

The time delay in the transmitter is the time which it takes the tape to move over an angle of 90 across the drum, and this is A; of the time in which the drum turns over 90, i.e., 7/8.50:0.0l74 sec. or 17.4 msec. This is very convenient, as pitch determining devices require a delay of about 12 msec. Thus by the time any part of the record has reached the centre of the window, the pitch can be correctly determined and the proper ring-zone selected.

In the receiver we must again have rings with scanning heads in the ratios 8:9: 10 15, hence the freedom in design is only whether we take the same numbers as in the transmitter, or multiples of it, say 16, 18 Let us take first the same numbers. As previously shown, the contact angle in the expander must be 64, which contains in the mean 1.42 heads in the ring with N :8. The repetition frequency must be again R=1OO cycles/sec, hence the expander, like the compressor drum, must rotate at 750 revs/min, and with a peripheral speed of 157"/sec. The speed of the tape must be of this, which is 22.4/ sec.

As the frequencies in the tape are 8-fold reduced, the minimum wave-length remains the same as in the transmitter, hence recording or scanning from the side opposite to the oxide coating is again justified. The delay now corresponds to the time in which the tape moves by 32, which is 0.071 sec. or 71 msec. If one wishes to reduce this delay, one must make the contact angle smaller,

which one can do, by the previously given rules, without violating the condition of correct reproduction of frequencies, by reducing the product w W in the same ratio as 6' The only disadvantage is that at the same time one reduces the number of heads simultaneously scanning the tape. If one wishes to reduce the delay of 71 msec. to 50 msec., the said number falls from 1.42 to unity, and with only one head scanning the tape at any one time there is some modulation noise, owing to the alternation of heads.

Consider now doubling the number of heads. The number covering a certain are or angle is now doubled, but at the same time the speed must be halved, hence the delay is also doubled. It is therefore most economical to have the same number of heads in the transmitter and in the receiver.

It can be seen that the relation between delay and number of heads simultaneously scanning the tape in the expander remains unaltered also if one increases the number of heads equally in the compressor and in the expander. If, for instance, one doubles both, the contact angles can be halved, but the speeds must also be halved, hence the delay remains the same. It is therefore necessary to strike a compromise between the delay and the quality of reproduction, which improves with the number of scanning heads simultaneously scanning the tape in the expander. One can however improve the quality by making the window-curve trapezoidal rather than elliptical, as shown in FIG. 8. If the window-curve were exactly trapezoidal, with a flat portion equal to one-half of its total length, there would be no change in the amplitude when one pick-up takes over from the other if their distance is /1 of the window width, that is to say if there are 1.33 pick-ups in the mean simultaneously scanning the tape. In order to realize .the optimum condition simultaneously for all head-rings one must make the flat portion of the trapezoid shorter for rings with a larger number of heads. This can be most easily achieved by spacing the heads in the recorder in a slightly non-uniform manner, so that the window-widths at the various headrings become somewhat unequal. Various other means for suppressing the periodic modulation of the signal, such as, e.g., a compensating periodic variation of the amplification factor, are well known in the electronic art.

Reference has been made to the compressing of speech signals for the purposes of transmission or recording.

Clearly the bandwdith compressed signals can either be transmitted over a line or radio link or can be recorded. In the latter event, instead of a receiver fitted with bandwidth expansion equipment the reproducer would be equipped.

I claim:

1. Apparatus for the processing of speech signals to produce a change in the bandwidth thereof comprising means for producing a plurality of separate records of the speech signals to be processed, a plurality of sets of pickup devices, each set comprising a plurality of pick-up devices and each set being associated with a different record, means for enabling said sets to scan the associated records at respectively different recurrence frequencies, means for combining the outputs of the pick-up devices of a set to produce a processed signal consisting of a changed bandwidth version of the speech signals, gating means in the outputfrom each set of pick-up devices, and means responsive to the pitch component of said speech signals for selectivelyoperating said gating means so as to continuously select one of said processed signals as an output signal.

2. Apparatus as claimed in claim 1 wherein each of said sets consists of a different number of individual pick-up devices.

3. Apparatus as claimed in claim 2 wherein each set of pick-up devices occupies a different circumferential zone of a rotatable drum and wherein the record is in the form of a plurality of tracks on a magnetic recording surface, a track being provided for each zone.

4. Apparatus as claimed in claim 3 wherein the magnetic recording surface is in the form of an endless magnetic tape arranged to be driven over the surface of said drum.

5. Apparatus for the processing of speech signals to produce a change of bandwidth comprising a magnetic recording surface having a plurality of parallel spaced apart recording tracks for producing a plurality of records of speech signals to be processed, a drum of conical form rotatable about its axis, a plurality of sets of pick-up devices provided on said drum, each set being located in a different circumferential zone thereof and being associated with a different record and each set having a different number of pick-up devices so that said sets operate to scan the associated records at respectively different recurrence frequencies, means for combining the outputs of the pick-up devices of a set to produce a processed signal consisting of an altered bandwidth version of the speech signals, gating means in the output from each set of pick-up devices, and means responsive to the pitch component of said speech signals for selectively operating said gating means so as to continuously select one of said processed signals as an output signal.

6. Apparatus as claimed in claim 5, wherein the pickup devices are in the form of gapped magnetic cores each provided with a winding, the windings of all the devices of each set being connected in series with each other and the windings of each set being connected to separate slip rings on the shaft of said drum.

7. Apparatus as claimed in claim 5 wherein the pickup devices are arranged in rings around a central cavity and magnetic recording tape and transporting means therefor are housed within said cavity, magnetic multitrack recording and erasing means associated with said tape also being housed within said cavity.

No references cited.

IRVING L. SRAGOW, Primary Examiner.

Claims (1)

1. APPARATUS FOR THE PROCESSING OF SPEECH SIGNALS TO PRODUCE A CHANGE IN THE BANDWIDTH THEREOF COMPRISING MEANS FOR PRODUCING A PLURALITY OF SEPARATE RECORDS OF THE SPEECH SIGNALS TO BE PROCESSED, A PLURALITY OF SETS OF PICKUP DEVICES, EACH SET COMPRISING A PLURALITY OF PICK-UP DEVICES AND EACH SET BEING ASSOCIATED WITH A DIFFERENT RECORD, MEANS FOR ENABLING SAID SETS TO SCAN THE ASSOCIATED RECORDS AT RESPECTIVELY DIFFERENT RECURRENCE FREQUENCIES, MEANS FOR COMBINING THE OUTPUTS OF THE PICK-UP DEVICES OF A SET TO PRODUCE A PROCESSED SIGNAL CONSISTING OF A CHANGED BANDWIDTH VERSION OF THE SPEECH SIGNALS, GATING MEANS IN THE OUTPUT FROM EACH SET OF PICK-UP DEVICES, AND MEANS RESPONSIVE TO THE PITCH COMPONENT OF SAID SPEECH SIGNALS FOR SELECTIVELY OPERATING SAID GATING MEANS SO AS TO CONTINUOUSLY SELECT ONE OF SAID PROCESSED SIGNALS AS AN OUTPUT SIGNAL.

Images