FIELD OF THE INVENTION

Our present invention relates to a method of and a system for handling conference calls among three or more telephone subscribers in a central office or exchange of digital type linked with these subscribers by pulse-code-modulation (PCM) channels.

BACKGROUND OF THE INVENTION

The digitization of voice samples in time-division-multiplex (TDM) telecommunication systems is well known. Thus, a code word of 12 bits -- including a polarity-indicating sign bit -- can be used to express 2¹¹ = 2048 distinct amplitude levels of a bipolar signal divided into eight ranges which are of identical widths on a logarithmic scale, i.e. a lowest range of 0- 15 units (e.g. millivolts), a second range of 16- 31 units, a third range of 32- 63 units etc., up to a highest range of 1024- 2047 units. As taught in commonly owned U.S. Pat. Nos. 3,755,808 and 3,789,392 issued in the name of Giampiero Candiani, such ordinary digital code words can be compressed into eight-bit codes and then re-expanded into their original 12 bits, with only minor loss of information, by expressing the rank of each code -- i.e. the order number of its amplitude range -- in binary form by means of three range bits which immediately follow the sign bit and are in turn succeeded by four quantum bits X, Y, Z, W representing the most significant bits following directly after the first "one" (or after the eighth "zero" in the case of the lowermost range) in the original code word. In more general terms, such a compressed code word consists in descending order of a sign or polarity bit, m range bits and n quantum bits.

A peculiarity of this type of compressed code word is the fact that the numerical weight of its quantum bits doubles from the second-lowest to the highest range while being the same in the two lowermost ranges. This prevents the direct positive or negative algebraic summing (i.e. adding or subtracting) of the range bits of two such codes, derived from a pair of coincident voice samples, unless they happen to be either of the same rank or of the two lowest ranks, referred to hereinafter as rank 0 and rank 1. Such algebraic summation is necessary in a conference call in which each participating subscriber is to receive superposed voice samples from all the other participants. It has therefore been the practice, in an exchange serving PCM channels of this type engaged in a conference call, to expand the incoming compressed codes for linear superposition and to recompress the resulting codes for transmission to PCM terminals serving the receiving subscribers. Such double conversion requires complex and correspondingly expensive central-office equipment.

OBJECT OF THE INVENTION

The object of our present invention, therefore, is to provide a simplified method of and system for linearly superposing significant portions of such logarithmically coded voice samples without expansion and recompression.

SUMMARY OF THE INVENTION

This object is realized, in accordance with our present invention, by separately storing a higher-ranking code and a lower-ranking code which are to be arithmetically combined, downshifting the quantum bits of the lower-ranking code to less significant bit positions in which they would have their original numerical weights if prefaced by the range bits of the higher-ranking code, comparing the polarity bits of the two codes, and subjecting the quantum bits of the higher-ranking code and the downshifted bits of the lower-ranking code to an algebraic summing operation of positive or negative sign, i.e. addition or subtraction, depending on the relative values of the corresponding polarity bits.

Since homologous quantum bits of codes of rank 0 and rank 1 have the same numerical weight, such a downshift will not occur if neither code has a rank higher than 1. Therefore, if the numerical difference of their ranks -- as expressed by their respective combinations of range bits -- has a value Δt, the downshifting will be by a number of bit positions equal to Δt except where the lower-ranking code is of rank 0; in the latter case, the number of bit positions involved in the shift will be Δt - 1.

By this downshifting operation we achieve a linearization of the quantum bits, i.e. of the significant bits X, Y, Z, W according to the above-identified Candiani patents, which are predominantly responsible for the transmission of voice frequencies and whose linear superposition is therefore essential for conference-type telephone communication. No such high fidelity is needed for the range bits, especially if one of two simultaneously speaking subscribers has a substantially louder voice than the other. It will generally be sufficient, as we shall demonstrate hereinafter, to disregard the range bits of the lower-ranking code in forming the algebraic sum and to preserve only those of the higher-ranking code in a composite code resulting from the summing operation, with possible incrementation or decrementation of their numerical value by one. With additive summing, we prefer increasing the range number by +1 in all instances except where both codes are of rank 0, along with a compensatory downshifting of the resultant quantum bits by one bit position. If the lower-ranking code has the rank 0, or if the ranks of the two codes are identical, this procedure is equivalent to linear addition of the entire codes. A positive carry produced by the adding of the quantum bits is accommodated in a bit position vacated by the last-mentioned downshift. With subtractive summing, conversely, a change in the range number of the higher-ranking code by -1 is necessary only when the calculation of the numerical difference of the two sets of quantum bits yields a negative carry; this decrease in the numerical value of the preserved range bits is to be compensated by a corresponding upshifting of the resultant quantum bits by one bit position, except in the case where the higher-ranking code is of rank 1 so that the downshift does not change the numerical weight of the quantum bits.

A system for carrying out this method comprises code-receiving means with two storage circuits, each including a first register for storing polarity and range bits and a second register for storing quantum bits of respective codes to be arithmetically combined, a first arithmetic unit connected to the first register for determining the difference in ranks between the two codes, linearization means connected to the second register and controlled by this arithmetic unit for carrying out the aforedescribed downshift of the quantum bits of the lower-ranking code, a second arithmetic unit connected to the second register and controlled by the contents of the first register for additive or subtractive summing of the two sets of quantum bits respectively held in the second registers of the two storage circuits, and evaluation means connected to the two arithmetic units for deriving a composite code from the range bits of the higher-ranking code and the quantum bits resulting from the algebraic summing operation. In the preferred embodiment or our invention more particularly described hereinafter, the evaluator includes a memory controlled by a program counter whose operation may be modified by associated logical circuitry under certain conditions such as, for example, the presence of two codes of rank 0 or the absence of a carry.

BRIEF DESCRIPTION OF THE DRAWING

The above and other features of our invention will now be described in detail with reference to the accompanying drawing in which:

FIG. 1 diagrammatically shows the general layout of a telephone network including a transit exchange designed to handle conference calls among three subscribers in accordance with our invention;

FIG. 2 is a block diagram of a conference circuit forming part of the transit exchange of FIG. 1;

FIGS. 3- 8 are more detailed diagrams of an input unit, a preprocessor, a first processor, a second processor, an output unit and an operating unit forming part of the conference circuit of FIG. 2;

FIG. 9 is a diagram of a transit exchange similar to that shown in FIG. 1 but designed for conference calls among four subscribers; and

FIG. 10 is a flow chart relating to the operation of a program counter included in the operating unit of FIG. 8.

SPECIFIC DESCRIPTION

FIG. 1 illustrates three telephone subscribers T₁, T₂ and T₃ assumed to be engaged in a conference call. For the sake of simplicity these subscribers have been shown as connected to a common terminal CT linked by outgoing and incoming PCM/TDM channels with a remote transit exchange TR, even though in practice such an exchange generally will not intervene in calls among subscribers served by the same terminal. Voice signals A, B and C respectively sent out by the three intercommunicating subscribers are converted in terminal CT, with the aid of a coder CC, into compressed eight-bit codes of the type discussed above and described in the two Candiani patents referred to. The coded signals are transmitted in respective channels (not necessarily consecutive, as indicated) of a PCM frame to a switching network RT of transit exchange TR which forwards them, in adjoining time slots, to a conference circuit CF where the codes are algebraically combined in conformity with our invention for retransmission, again via network RT, to terminal CT in combinations B+C, A+C and A+B in channels assigned to them by the network. A decoder ED reconverts these composite eight-bit compressed codes into analog signals and sends them on to their respective destinations, i.e. signals B+C to subscriber T₁, signals A+C to subscriber T₂ and signals A+B to subscriber T₃. The compounding of each pair of codes in circuit CF takes place in the course of an eight-bit time slot so that, in principle, all the subscribers using the available PCM channels could participate simultaneously in three-way or possibly four-way conference calls.

In FIG. 2 we have shown the conference circuit CF of FIG. 1 as comprising an input unit UI receiving incoming codes on a lead label "in", and working into a preprocessor UP feeding range bits to a first processor UT and quantum bits to a second processor US. The two processors deliver the results of their operations to an output unit UU which periodically discharges the composite codes via a lead labeled "out". The three processing units UP, UT and US form part of an evaluation section which also includes an operating unit UO coacting therewith.

Input unit UI, illustrated more fully in FIG. 3, is under the control of a timer including a frequency divider DF to which clock pulses CK₄ are fed from a nonillustrated generator with a cadence equaling twice the bit rate. Divider DF halves this cadence so as to produce a train of pulses CK₂ which step an 8-bit pulse counter CN₁ reaching its full count with the arrival of the last bit of the code word in a time slot of a periodically recurring PCM frame assumed to contain 32 such time slots, each recurrent time slot constituting a TDM transmission channel as is well known per se. Another pulse counter CN₂ of 32-pulse capacity is stepped by the counter CN₁ at the beginning of each new time slot to count the 32 channels. The several stage outputs of counter CN₂ are connected to a read-only memory RI which energizes a different output lead during each time slot of a frame, only six output leads z₁ -- z₆ having been illustrated.

The bits of a code successively arriving over input lead "in" are fed in parallel to a number of registers of which only three, labeled RS₁, RS₂ and RS₃, have been illustrated. These registers are provided with enabling inputs including respective AND gates N₁, N₂, N₃ each receiving the clock pulses CK₂. The other inputs of these AND gates are respectively connected to output leads z₁, z₂ and z₃ of memory RI whereby only one register receives the clock pulses CK₂ in the course of a given time slot. In the case here assumed, register RS₁ is serially loaded in the first time slot of a frame with the coded voice sample A (see FIG. 1), registers RS₂ and RS₃ respectively receiving the codes B and C during the next-following two time slots. In response to a start pulse SP emitted by counter CN₁ upon the completion of its count, registers RS₁, RS₂ and RS₃ are discharged in parallel over respective 8-lead multiples into associated registers RS'₁, RS'₂ and RS'₃, respectively, so as to become available for the storage of new codes in the following cycle. The bits stored in ancillary registers RS'₁, RS'₂ and RS'₃ are serially read out, under the control of memory outputs z₄ - z₆ as described below, via leads y₁, y₂ and y₃ extending to a pair of multiplexers MX₁ and MX₂ forming part of preprocessor UP as shown in FIG. 4.

Multiplexer MX₁ has two inputs connected to output lead y₁ of register RS'₁ and a third input connected to output lead y₂ of register RS'₂. Multiplexer MX₂ has two inputs connected to output lead y₃ of register RS'₃ and a third input connected to output lead y₂ of register RS'₂. Both multiplexers are simultaneously switchable by signals on leads z₄, z₅ and z₆ to direct the bits of a selected pair of stored codes to their respective outputs, i.e. codes B and C upon the energization of lead z₄, codes A and C upon the energization of lead z₂ and codes A and B upon the energization of lead z₆. The following discussion will be limited to the period of energization of lead z₆, this being the time slot during which the composite code A + B is to be produced for transmission via PCM terminal CT to subscriber T₃ as described above with reference to FIG. 1.

The bits of codes A and B are thus concurrently read out, in sequence, from multiplexers MX₁ and MX₂ into respective branches of preprocessor UP, i.e. a first branch including two registers T_a, H_a and a second branch including two registers T_b, H_b. Registers T_a and T_b are of 4-bit capacity designed to accommodate the polarity bit S_a or S_b and the three range bits of their respective codes, collectively designated t_a, t_b . Registers H_a and H_b also have four stages each, designed to receive the quantum bits of their respective codes, collectively designated h_a, h_b, as well as a fifth stage for the storage of a supplemental bit f_a, f_b indicating by its presence that the corresponding code is of a rank higher than zero. This fifth stage is shown only for convenience as part of register H _a or H_b but may in fact be considered a separate 1-bit store. The contents of the other four stages, containing the quantum bits, are shiftable to the right under the control of respective signals K₁ or K₂ for the purpose and under the conditions described hereinafter, such a shift occurring upon the occurrence of basic clock pulses CK₄. These four stages can also be immediately emptied by means of clearing pulses K₁ ' and K₂ ', respectively.

The writing inputs of registers T_a and T_b are connected directly to the outputs of the associated multiplexers MX₁ and MX₂ whereas the writing inputs of registers H_a and H_b are connected to the same outputs by way of further multiplexers MX₃ and MX₄. The loading of these registers is controlled by timing signals issuing from a read-only memory RP in unit UO controlled by a program counter CP as shown in FIG. 8. More specifically, registers T_a and T_b are loaded under the control of a signal c₁ whereas registers H_a and H_b are loaded under the control of a subsequent signal c₃ in the presence of a switching signal c₂ controlling the multiplexers MX₃ and MX₄. The range bits t_a and t_b appearing in three of the four stage outputs of registers t_a and t_b are fed to three comparators ID₁, ID₂ and ID₃ serving to identify the respective codes A and B in terms of relative rank. Comparators ID₁ and ID₂ have a threshold level 0 so as to emit respective signals t_a > 0 and t_b > 0 if these codes are of other than rank -0. These two signals are applied to respective inputs of a pair of AND gates N₄ and N₅ also receiving a control signal c₅ from memory RP of FIG. 8; AND gates N₄ and N₅ are connectable via multiplexers MX₃ and MX₄ upon the disappearance of signal c₂, to the writing inputs of registers H_a and H_b to store therein the bits f_a and f_b if the respective comparators ID₁ and ID₂ have an output. The third comparator ID₃ receives both sets of range bits t_a and t_b to energize one of three output leads with a respective signal t_a > t_b, t_a = t_b or t_a < t_b, collectively designated t. A fourth comparator ID₄ receives the two sets of quantum bits h_a and h_b from registers H_a and H_b, energizing one of three output leads with a respective signal h_a > h_b, h_a = h_b or h_a < h_b, collectively designated h.

Processor UT, shown in FIG. 5, comprises two multiplexers MX₅ and MX₆ receiving the range bits t_a and t_b in parallel from registers T_a and T_b. These multiplexers, whose inputs are simultaneously switchable by a signal f_c generated by a logic network RL, work into respective inputs x₁ and x₂ of an arithmetic unit AU. Logic network RL receives the output signals t_a > 0, t_b > 0. t and h from comparators ID₁ and ID₄ as well as a control signal c₄ from memory RP (FIG. 8) and the output of arithmetic unit AU which is either the numerical difference Δt of the bit combinations fed to its inputs x₁ and x₂ or only the numerical value t* of the combination of range bits at input x₁, depending on the absence or presence of a timing signal c₆ delivered to an inverting input of an AND gate N₁₀ which is interposed between multiplexers MX₆ and input x₂. Switching signal f_c comes into existence whenever the combination of bits t_a and h_a has a numerical value larger than that of the combination of bits t_b and h_b, i.e. when the absolute magnitude of the coded signal A exceeds that of the coded signal B. Thus,

fc = (ta > tb) + (ta = tb) · (ha > hb)                                                  (1)

with f_c = 1, bit groups t_a and t_b are respectively supplied to inputs x₁ and x₂ ; with f_c = 0 the converse is the case. Unit AU, accordingly, always received the range bits of the code of larger absolute magnitude on its input x₁ and those of the code of smaller absolute magnitude on its input x₂. If the two codes happen to be of equal magnitude, the position of the multiplexers will be immaterial; in that case, as will be described hereinafter, the arithmetic unit AU does not intervene in this operating phase in which c₆ = 0.

The values Δt and t* issuing in different operating phases from unit AU are loaded, under the control of a signal c₇ from memory RP, into a reversible counter CS designed to be stepped backward by clock pulse CK₄ in the presence of signal c₄ applied to a countdown input thereof. Counter CS also has a "subtract 1" input, energizable by an output signal a₈ of a logic network LR of unit UO (FIG. 8), and an "add 1" input, energizable via an AND gate N₆ by a control signal c₁₃ from memory RP in the presence of a signal i_s indicating identity of the polarity bits s_a, s_b as determined by an Exclusive-OR gate EO in unit US (FIG. 6). Counter CS has three outputs delivering range bits Q₂, Q₃ and Q₄ to an eight-bit final register RC of output unit UU shown in FIG. 7. The three counter outputs are further connected to logic network RL and to three logic gates in parallel, namely a NOR gate O₁, an AND gate N₈ with two inverting inputs for bits Q₂ and Q₃, and an AND gate N₉ without inverting inputs. NOR gate O₁ generates a signal

i0 = Q2 · Q3 · Q4    (2)

whenever the count of component CS is 0. AND gate N₈ generates a signal

i1 = Q2 · Q3 · Q4    (3)

when the count has a value of 1. AND gate N₉ generates a signal

i7 = Q2 · Q3 · Q4    (4)

in the presence of a maximum count of numerical value 7.

Signals K₁ and K₂, designed to cause a downshift of the quantum bitsh_a and h_b in registers H_a and H_b of FIG. 4, can be expressed by

K1  = c4 · (ta < tb) ·(ta = 0) ·(Δt =1)                                   (5)

and

K2  = c4 ·(ta > tb) ·(tb =0) ·(Δt = 1)                                  (6)

clearing signals K₁ ' and K₂ ' can be written as

k1 ' = c4 ·(tb > ta) · (Δt > 4) (7)

and

K2 ' = c4 · (tb < ta) · (Δt > 4)                                                        (8)

Equations (5) and (6) indicate that the register involved in the downshift is the one which carries the quantum bits of the smaller of the two codes A and B, i.e. register H_a in the case of t_a < t_b and H_b in the case of t_b < t_a. These equations also show that no shift occurs, if the smaller code is of rank 0 and the difference Δt between the two ranks is 1, i.e. if the larger code is of rank 1. Since the extent of the downshift is determined by the numerical differnce Δt, as will be more fully described hereinafter, a shift of the four quantum bits by more than three bit positions would completely empty the register stages assigned to these quantum bits. It is for this reason that clearing signals K₁ ' and K₂ ' will generated under the conditions indicated by equations (7) and (8), i.e. with the numerical difference Δt exceeding the value 4.

Unit US, shown in FIG. 6, comprises two multiplexers MX₇ and MX₈ which are analogous to multiplexers MX₅ and MX₆ of unit UT and are also controlled by the switching signal f_c. These multiplexers receive the quantum bits h_a and h_b from registers H_a and H_b (FIG. 4) and deliver them in parallel to respective inputs x₂ ' and x₂ ' of an arithmetic unit AU'. Again, the bits of the larger code (as determined by signal f_c) are invariably applied to input x₁ ' whereas those of the smaller code go to input x₂ '. Unit AU' performs either an addition or a subtraction, depending on the presence or absence of the polarity-discriminating signal i_s issuing from XOR gate EO. Gate EO has an output i_s = 1 only if bits s_a and s_b have different values, indicating opposite polarities of the voice samples represented by codes A and B. Bits s_a and s_b are available at the first stage outputs of registers T_a and T_b of preprocessor UP (FIG. 4). In the case of such diversity the quantum bits h_a and h_b must be summed differentially, rather than additively as when the two samples are of like polarity. Unit AU' works through an OR gate O₂ into a shift register RU with a storage capacity of four bits. Under certain conditions giving rise to a signal a₉ in the output of logic network LR (FIG. 8), more fully described hereinafter, register RU can be loaded with an all-1 bit combination from a store ST via OR gate O₂. The bits stored in register RU can be shifted either to the left (up) or to the right (down) under the control of respective signals a₇ and c₁₃. An upshift can be inhibited by a signal a₅ and the entire register can be reset to zero by a clearing signal a₂. Signals a₂, a₅ and a₇ are generated by the logic network LR of FIG. 8 under conditions to be described. A carry signal f₀, originating at a read-only memory RZ of unit UO (FIG. 8), can be used to insert a bit into the first stage (at extreme left) of register RU vacated by a downshift. The register has four stage outputs carrying the quantun bits Q₅, Q₆, Q₇ and Q₈ of the composite code corresponding to the bits designated X, Y, Z and W of the above-identified CANDIANI patents.

Register RC of output unit UU, illustrated in FIG. 7, has its last four stage inputs connected to the stage outputs of register RU to receive the bits Q₅ -Q₈ therefrom, together with the bits Q₂ -Q₄ from the counter CS of unit UT (FIG. 5). A multiplexer MX₉ controlled by a switching signal f_c delivers either the polarity bit s_a or the polarity bit s_b to the first stage of register RC to form the corresponding bit Q_s of the outgoing composite code, depending on whether code A or code B has the larger absolute value.

As shown in FIG. 8, the operational unit UO includes a logic section UL in addition to a program counter CP in memory RP. Logic section UL comprises the aforementioned network LR and memory RZ along with two multiplexers MX₁₀ and MX₁₁ feeding the bits f_a and f_b from the unshiftable sections of registers H_a and H_b to respective inputs of memory RZ under the control of switching signal f_c. Thus, the presence of a signal a* in the output of multiplexer MX₁₀ indicates that the larger of the two codes A and B has a rank greater than zero; similarly, the signal b* is present in the output of multiplexer MX₁₁ if the smaller code is also at least of rank 1. Signal a*, whose absence represents the condition t_a = t_b = 0, is also fed to the logic network LR along with timing signals c₄ and c₈ -c₁₁ from memory RP, output signals i₀, i₁ and i₇ of counter CS (FIG. 5), signal Δt from arithmetic unit AU and signals t, h, from comparators ID₃, ID₄ of FIG. 4. Network LR generates the various command signals a₁ -a₉ partly referred to above. Program counter CP has an operating cycle of 16 phases representing as many cycles of the basic clock signal CK₄. In phase φ₀ its program is arrested and the contents of register RC (FIG. 7) are read out. The program is started by the pulse SP (see FIG. 3) and can be foreshortened by the de-energization of a reset input normally receiving voltage from an AND gate N₇ with inputs receiving signals a₂, c₁₀ and a₄. A "hold" input arrests the program in the absence of a signal a₃. Two other inputs, normally receiving signals a₁ and a₆, cause a jump of the program to phase φ₁₁ or φ₁₅ upon the disappearance of these signals in phases φ₉ and φ₁₁, respectively.

The sequence in which the output signals c₁ -c₁₃ are generated by memory RP, in the various phases of program counter CP, has been indicated in the flow chart of FIG. 9.

In the first four phases, φ₁ -φ₄, signal c₁ is present and causes the loading of registers T_a and T_b with polarity bits s_a and s_b and range bits t_a, t_b, respectively. In the next four phases φ₅ -φ₈, signals c₂ and c₃ enter the quantum bits h_a and h_b in the registers H_a, H_b. In phase φ₉ the multiplexers MX₃ and MX₄ are switched by the disappearance of signal c₂ ; with signal c₅ generated and with registers H_a and H_b still enabled by signal c₃, these multiplexers write the bits f_a and f_b in the first stages of these registers, displacing the quantum signals already there by one bit position. Signal c₉ is also generated at this point.

With the appearance of signal c₉, logic LR explores the possibility that the two codes to be combined are of equal rank. If this happens to be the case, the command a₁ is generated according to the formula

a1 = c9 · (ta = tb)           (9)

Signal a₁ causes a jump to state φ₁₁ unless overridden by a signal a₂ according to the formula

a2 = c9 · (ta = tb) · (ha = hf) · is                               (10)

indicating that the two samples are of the same amplitude but opposite sign. The occurrence of signal a₂ resets the program counter CP to its state φ₀ whereby register RC of input unit UU (FIG. 7) is unloaded with an all-0 code since the same signal a₂ resets the register RU of FIG. 6 (state φ₉ ' in FIG. 10) and since the counter CS is empty at this instant as will presently become apparent.

If t_a ≠ t_b, signal c₇ generated in phase φ₁₀ causes the unit CS to be loaded with the difference Δt computed in the interim by the arithmetic unit AU. If Δt > 4, clearing signal K₁ ' or K₂ ' is generated according to equation (7) or (8) to empty the register H_a or H_b of quantum bits as already described (state φ₁₀ ' ).

If the value of Δt is > 1 but does not exceed 4, signal a₃ disappears according to the formula

a3 = i1 · (Δt < 4) · c4 (11)

under the control of signal c₄ generated in this phase. Program counter CP is thus put on "hold" for as many clock cycles as is necessary to reduce the count of component CS to 1. At the same time the quantum bits in register H_a or H_b (whichever happens to carry part of the larger code) are downshifted by as many positions by signal K₁ or K₂ until the program is restarted by the appearance of signal a₃ with i₁ = 1. At this instant the logic network RL receives from counter CS an output signal equivalent to Δt = 1 so that signal K₁ or K₂ terminates if the smaller code happens to be of rank 0, i.e. if t_b = 0 or t_a = 0 as the case may be. In that instance the downshifting of register H_a or H_b stops but the counter CS takes one further step to 0. If, however, the smaller code has a rank higher than 0, the downshift goes on until the disappearance of signal c₄ which may occur in the 11th, 12th, 13th or 14th phase according to the number of clock signals required to step down the counter CS. The program counter CP then reaches the state φ₁₁ which therefore does not necessarily coincide with the 11th phase.

In state φ₁₁ the signals c₆ and c₈ come into existence while signal c₇ recurs. Signal c₆ blocks the input x₂ of arithmetic unit AU so that only the range bits of the larger codes are received by unit AU at its input x₁ and are passed through to encounter CS. The following alternatives will now have to be considered:

Polarity-discriminating signal i_s may be either 0 or 1. In the first instance, the addition of the quantum bits present on inputs x₁ ', and x₂ ' of arithmetic unit AU' (FIG. 6) may or may not give rise to a positive carry. In the second instance, since the numerical value of the quantum bits of the larger code may be less than that of the quantum bit of the smaller code, the subtraction performed in the unit AU' may result in a negative carry. The occurrence of either type of carry is marked by an output signal r of unit AU' delivered together with signal i_s to memory RZ of operating unit UO. The carry signal f₀ is given by:

f0 =  is · r · (a* + i · b*) + is · r                                              (12)

Thus, in the case of an adding operation (i_s = 0) the presence of a signal r will produce the signal f₀ only if at least one of the two codes has a rank higher than 1. In the case of subtraction, signal f₀ will be generated as long as signal r is absent.

With the occurrence of signal c₈ in state φ₁₁ the signal a₄ will disappear if no carry is present and if at least one of the codes has the rank 0, according to the formula:

a4 =  c8 · (is · r · b* + is · f0 + is · a* +  is · r · i7 + is ·  i7 · b*) (13)

This immediately resets the program counter PC and fills the register RC with the bits held in counter CS and register RU.

If, in the additive case, a carry is present and/or the smaller code is of a rank greater than 0, signal a₉ comes into being according to the formula

a9 = is · i7 · (r + b*)   (14)

This loads the register RU with a bit combination 1 - 1 - 1 - 1 so that an all-1 code is delivered to register RC.

If the aforedescribed conditions are not met, program counter CP continues to its next state φ₁₂ in which signals c₁₀ and c₁₃ are generated. Signal c₁₃ now increases the count of component CS by 1 and simultaneously downshifts the contents of register RU by one bit position unless such shift is inhibited by a signal a₅ according to the formula :

a5 = i0 · c10                      (15)

indicating that the larger code is of rank 0. Signal c₁₀ then steps the counter CP to its initial state φ₀ while a bit corresponding to signal f₀ is introduced into the stage of register RU vacated by the 1-position downshift caused by signal c₁₃.

If the two voice samples are of opposite polarities, signal a₆ disappears in state φ₁₁ according to the formula:

a6 = is · c8                       (16)

and causes a jump to state φ₁₅ with generation of signal c₁₁, producing in turn the signal a₈ given by

a8 = is · c11                      (17)

Signal a₈ diminishes the count of component CS by 1 and is accompanied by a 1-position upshift of the contents of register RU by 1-bit position under the control of signal a₇ as long as counter CS has an output other than i₇, according to the formula:

a7 = i1 · c11                      (18)

From state φ₁₅ the program counter reaches a final state φ₀ to stop the program.

Since equation (13) includes the terms of equation (14), a resetting of the program counter occurs also under the conditions in which an all-1 code is entered in register RC. Counter CS is cleared during each such resetting.

              EXAMPLES______________________________________                        NumericalCode                         Value______________________________________I         000    -     1100 Δt                           =   0    12+II       000    -     1000 r   =   1    8I+II      001    -     0100 f0                           =   0    20I         001    -     1100 Δt                           =   1    28+II       000    -     1000 r   =   1    8                  0100 f0                           =   0DOWNSHIFT              0010I+II      010    -     0010              36I         010    -     1000 Δt                           =   2    48+II       000    -     0100 r   =   0    4                  0100 f0                           =   0DOWNSHIFT              0010I+II      010    -     1010              52I         010    -     1110 Δt                           =   0    60+II       101    -     1100 r   =   1    56                  1010 f0                           =   1DOWNSHIFT              0101I+II      011    -     1101              116I         010    -     0001 Δt                           =   2    34-II       000    -     0100 r   =   1    -4DOWNSHIFT              0010 f0                           =   0                  1111UPSHIFT                1110I -II     001    -     1110              30______________________________________

In FIG. 9 we have shown a modification of exchange TR to handle a conference call among four subscribers sending out voice codes A, B, C and D in respective time slots 101-104 and receiving compound codes B+C+D, A+C+D, A+B+D and A+B+C in respective time slots 201-204. Within exchange TR the codes are transmitted from network RT to conference unit CF in six time slots 301-306 and, after compounding, returned to the network in six time slots 401-406. The codes A and B alternate in time slot 306 even as codes C and D alternate in time slot 303. Thus, these codes may be considered as grouped in two conference combinations cf₁ and cf₂. The compound codes A+B and C+D, occurring in time slots 403 and 406, are not utilized by the switching network RT.