Interpolation of data with continuous speech signals

BACKGROUND OF THE INVENTION

This invention relates to communication systems and, more particularly, to multiplex transmission systems in which both voice communication and data signals are transmitted over a single communication channel.

Field of the invention

In some communications systems, the need arises for transmitting speech signals and coded data over a single transmission channel. Multiplex systems permit this to be done. They generally have followed one of two patterns; either an allocation of signals to selected frequencies or to selected times. If a restricted bandwidth service is adequate, the available transmission band may thus be allocated on the frequency scale for two or more separate signal trains. Simultaneous data and speech transmission may thus take place continuously with each signal necessarily restricted to only a portion of the available channel bandwidth. Alternatively, if full-time service is not necessary, data signals may be interspersed during gaps in the speech, either in natural silent intervals or in gaps created by chopping. The first technique, though permitting simultaneous, continuous service, limits the quality and quantity of data and speech that can be transmitted. The second, though permitting full bandwidth transmission, is discontinuous and may severely degrade the quality of the speech signal.

Description of the prior art

Transmission systems in which data or signaling information is interspersed in a continuous speech signal are effective because a normal speech signal contains many gaps and pauses as well as listening intervals completely devoid of speech. Data may be transmitted during these gaps and pauses.

To increase the flow of data, it has been proposed that the speech signal be interrupted for the insertion of data at other than natural moments of speech silence. If the required gaps for data are created simply by interrupting the speech signal instead of waiting for intervals of silence, severe degradation of speech quality and some loss in intelligibility results. Of course, if this form of degradation can be tolerated, or is held to a low level, a greater allocation of data to the channel may be made.

The primary reason for the speech degradation which results from speech interruption or chopping is twofold:

3,492,429 fashion well known to those skilled in the art, the quotient s(f)/a(i). Envelope detector 14 yields at its output the envelope of s(t), viz., α(ί). The envelope signal is passed through low-pass filter 15 to produce a constant signal c=aa(t). The constant signal c is thereupon subtracted from envelope signal a(t) in subtractor 16 to produce a difference signal a(.t)—c, which is delivered both to half-wave rectifier network 17 and threshold network 18. Half-wave rectification of the quotient signal ₁₀ produces the desired switching function { }₊. A typical switching signal is illustrated in FIG. 4 (δ). The signal is used to control the action of switch modulator 19, supplied at its signal input with the quotient signal s(t)/a(i). The resulting product is equal to the desired switching 15 function s,(i) as defined in Equation 4 and illustrated in FIG. 4(c). It is reduced to baseband by application to single sideband modulator 20 which is supplied with carrier signals at frequency f_c from oscillator 12. The resulting discontinuous speech signal, ²⁰ {«(/)— c}₊ cos Ev(t)] (8) is applied to one input of adder 21.

Data signals J, of any desired sort are applied to the terminal apparatus of FIGURE 1, for example at ter25 minal 22. Because of the discontinuous nature of their transmission, the applied data signals are passed through data buffer 23. In typical fashion, data buffer 23 absorbs applied continuous signals, stores them, and makes them available upon call at irregular times. The transfer of data 30 signals from the buffer to modulator 24 is under control of signals supplied from threshold network 18. Thus, data is transferred out of the buffer only when the different signal α(ί)—c is below a selected ratio factor. The threshold of network 18 is thus set in order that con35 trol signals reach buffer 23 when the envelope of the continuous input signal is arbitrarily small and at pitch synchronous chopping points. Data signals from buffer 23 are thus delivered to modulator 24. Modulator 24 is supplied with carrier signals at frequency f₀ from oscillator 40 25, to produce a modulated output signal, di cos (2ττ/₀ί).

The modulated signal is supplied to the second input terminal of adder 21, and there interleaved in the interrupted speech signal supplied from modulator 20. Since the interruptions in the speech occur synchronously with 45 the delivery of data, a substantially continuous output signal is produced. The composite output signal may then be transmitted from terminal 26 in any desired fashion to a receiver station.

A suitable receiver terminal for composite data and 50 speech signals is shown in FIGURE 2. Composite signals received at terminal 30 are examined in single frequency detector 31 for carrier signals at frequency /₀· The presence of this signal indicates an interval of data transmission. Accordingly, switch 32 is actuated to de55 liver the composite input signal from terminal 30 to envelope detector 33. The detector, which may be of any conventional type, removes the carrier f₀ and delivers a baseband data signal to data buffer 34. In conventional fashion, data is accumulated in the buffer and, as desired, 60 delivered as a continuous data output signal, for example, to terminal 35, and thereupon to a data utilization device, not shown.

In the absence of a detected signal at frequency /o> switch 32 supplies input signals from terminal 30 dies rectly to speech output terminal 36 for delivery to a ³

FIG. 3 is a block schematic diagram of a data signal detector which may be used in the receiver apparatus of FIG. 2; and

FIG. 4 is a waveform diagram illustrating the development of a suitable switching function in accordance with the invention.

DETAILED DESCRIPTION

A pitch synchronous switching function for optimumly interrupting a speech signal for the insertion of data is developed in accordance with the invention by establishing a selected threshold for speech signals and pitch synchronously interrupting the signal when its amplitude falls below the established threshold. Thus, the average offtime ratio of the speech signal is a function of the established threshold. Regardless of the ratio, however, interruption takes place at prescribed instants such that discontinuities in the remaining speech signal are harmonically related, thus minimizing audible degradation.

The switching function is derived in the following manner. An applied speech signal, s(t)=a(f) cos [φ(ί)] (1) is divided by its envelope, _fl(f) = [s2(Z)+S2(_i)_]l/2_j (2) where s(t) is the Hilbert transform of j(i). The resultant quotient signal is thereupon multiplied by a modified envelope signal derived from the original envelope a(t). The modified envelope signal is developed from the original envelope by subtracting from it a constant proportional to the envelope and subjecting the difference to half-wave rectification. The constant c subtracted from the envelope a(f) determines the extent of the gaps in the speech signal s(i); it determines the threshold of switching and the off-time ratio. The modified envelope signal, «(/) = {a(t)-c}₊ (3) =a(t) — c if a>c, =0 if a<c, where the function { }₊ equals its argument for positive arguments and is zero otherwise. The product of the quotient signal s(t}/a(t) when multiplied by the modified envelope d(t) yields the desired interruption signal, (4)

As noted, the average off-time ratio depends on the magnitude of the constant c. For Gaussian signals, the ratio is given by r_off=l— exp[—c²/a²](5)

Typically, c is made equal to 0.5α. For a Gaussian signal, this choice corresponds to ϋ=1/4[πά²]¹''²(6)

Thus, the average off-time becomes r_oil=l—exp[—tt/16]=0. 18=18%(7)

Actual observed off-time ratios for c=0.5a 'are approximately 26% for male speech and 16% for female speech. In this observation, a was obtained by averaging α(ί) over 20 msec, with a rectangular time window.

FIG. 1 illustrates apparatus for carrying out the operations defined mathematically above. A continuous speech signal s(i) for example, of the form illustrated in FIG. 4(a), is applied to the transmitter apparatus of FIG. 1, for example, at input terminal 10. For implementation simplicity, the signal is modulated on a carrier frequency, fc> by passing it through single sideband modulator 11 which is supplied with carrier signal oscillations at frequency f_c from oscillator 12. The resulting modulated signal is thereupon delivered both to infinite clipper 13 and to envelope detector 14. The single sideband modulation step together with infinite clipping yields, in a speech utilization device, not shown. Although output signals available at terminal 36 are interrupted, the interruptions, as noted above, are insignificant insofar as speech intelligibility is concerned since the discontinuities 76 bounding the gaps are small and harmonically related.

Hence, the speech signal may be used directly with minimum perceivable degradation.

A suitable single frequency detector for use in the receiver apparatus of FIGURE 2 is shown in FIGURE 3.

The composite input signal is passed through baseband

3,492,429 filter 40, centered about frequency /₀, in order that a signal component at frequency f₀, if present, is delivered to threshold network 41. Network 41 responds only to signals above a selected threshold Θ to produce a control signal. Thus, if a component at frequency f₀ of substantial magnitude is present, indicating that a carrier signal from oscillator 25 at the transmitter is present in the composite signal, as control signal is produced by network 41 and delivered to logic gate 42. The composite signal is also supplied to notch filter 43 centered about frequency f₀ in order to block components at that frequency. Thus, if a speech or noise signal component at frequency f_ohappens to be present in the composite signal (so as to reach threshold network 41), it is blocked by filter network 43. However, other speech or noise components which accompany the component at f_o are passed by filter 43 and, if of sufficient magnitude, exceed the selected threshold of network 44. A control signal is thereupon delivered by network 44 to logic gate 42. Thus, network 44 provides an additional control signal for logic gate 20 42 to improve detection reliability. If the network of FIGURE 3 detects a carrier signal at frequency f₀, logic gate 42 supplies the appropriate control signal to switch 32 (FIGURE 2) for routing incoming signals to the data channel. If the network of FIGURE 3 fails to detect a 25 carrier component, the signal from gate 42 actuates switch 32 to connect the input terminal to the speech channel.

The above-described arrangements are, of course, merely illustrative of the application of the principles of 30 the invention. Numerous other arrangements may be devised by those skilled in the art without departing from the spirit and scope of the invention.