The invention relates generally to analog-to-digital and digital-to-analog conversion systems associated with proper scaling before A/D conversion and after D/A conversion, and particularly to such a system for an echo canceller.

Analog-to-digital and digital-to-analog converters have come widely into use because of the development of digital processing systems. For instance, in the technical field of control and monitoring systems, that is devices for maintaining conditions in operating systems as close as possible to desired values despite changes in the operating environment, analog components have traditionally been used. In the 1970's, the use of discrete or logical control elements and programmable logic controllers became widespread, and allowed the development of digital control system for monitoring such things as chemical processes, machine tools, other industrial processes and operations. To achieve this goal, analog-to-digital A/D converters are used to transform analog information, such as audio signals or measurements of physical variables (for example, temperature, force, or electric voltage) into a form suitable for digital handling. Figure 1A illustrates a conventional case of process control involving analog components only, in which the use of negative feedback provided by block 3 produces changes in the characteristics of the system which improve the performance of the system. In this automatic control system, feedback is used to compare by means of subtracter 1, the actual output S of a system with a desired input E, the difference appearing at the output of subtracter 1 being used as the input signal of a controller 2. A high performance of the system in terms of dynamic performance and stability, often involves a sophisticated feedback function in block 3 which may advantageously be designed by means of precise and cheap digital processing systems, as in figure 1B. However, when both output S and input E of the system are analog variables, the use of digital technology for providing feedback function block 4 involves the conversion of the first variable into digital form prior to any computation, and then the conversion of the digital output of block 4 into analog form in order to generate the analog variable E&min; that will be eventually used to produce the difference E-E&min;. However, in some environment of changing characteristics, the level of the electric voltage S to be converted is likely to vary significantly in a wide range. It is therefore essential to perform a proper scaling of the analog signal S prior to its conversion, in order to make the best use of the A/D converter precision. This A/D conversion is therefore performed after a amplification step in block 6. After computation in block 4, the digital result is converted back into its analog form in block 5 which performs the D/A conversion and then an attenuation for providing the analog value E&min;. In order to allow a high performance of the system, the precise mastering of the transfer functions of block 4, 5 and 6 is most desirable and particularly, both transfer function of blocks 5 and 6, i.e. both A/D-amplification and D/A-attenuation processings, should be accurately inverse of each other.

Similarly, echo cancellation techniques allowing high speed full-duplex data communication on a single channel also require precise transfer functions included into the echo cancelling loop in order to achieve high rejections ratios for the echo. The cancellation of the echo is achieved by an echo estimator which generates an estimation of the value of the echo signal that spoils the received signal from a far-end data control equipment such as a modem. The estimated echo is subtracted to the received signal in order to produce a signal being as close as possible to the ideal received signal having no echo. The estimation of the echo by means of digital processing systems in a standard analog 4-wires modem again involves two accurate A/D-amplification and D/A-attenuation transfer functions. This particular case will be described in detail with respect to figure 7A, 7B and the following figures.

In the background art, the design of precise transfer functions, and particularly two transfer functions being accurately inverse of one another for performing a A/D conversion associated with an amplification step, and a D/A conversion associated with a attenuation step have always involved adjustable components, precise elements which inevitably increase the final cost of the system. Moreover, even precise adjustments can not guarantee accurate transfer functions since the values of the components are subject to long-term shift.

Therefore, it is an object of the invention to provide two transfer functions being accurately inverse of one another, the first one for performing an A/D conversion associated with an amplification step, and the second for performing a D/A conversion associated with an attenuation step.

It is another object of the invention to provide a conversion system for performing either an A/D-amplification conversion or a D/A-attenuation conversion whereby allowing easy use of digital processing systems in conventional analog control systems.

It is further object of the invention to provide a conversion system for performing either a A/D-amplification conversion or either a D/A-attenuation conversion whereby allowing the design of a cheap and high performance echo cancellation device for a data control equipment.

In order to attain the above objects, there is provided a conversion system for performing either an analog-to-digital A/D conversion associated with an amplification step or either a digital-to-analog D/A conversion associated with an attenuation step. The system includes means (115) for receiving an input digital word to be processed, i.e. converted into analog and then attenuated, and means (165) for receiving an input analog value to be processed, i.e. amplified for scaling purpose and then converted into digital. It also includes an digital-to-analog D/A converter (110), an attenuator (120) for attenuating the analog output of D/A converter (110), and a comparator (150) for comparing the value of the input analog value to be processed and the output of said attenuator (120). The processing of the D/A-attenuation process is performed by means of both D/A converter (110) and attenuator (120). In order to achieve the A/D-amplification process, the system further includes means (140) for generating a sequence of digital words to the D/A converter (110), and means (220) for storing among this sequence, the digital value that minimizes the difference between both inputs of comparator (150). This digital value is extracted as being the digital representation of the amplified analog input value. Since both A/D-amplification and D/A-attenuation processings involve the same physical components, both processing have transfer function exactly inverse one another. The typical use of this circuit is in echo cancellation technique.

Fig. 1A and 1B illustrate the use of a digital feedback function in an analog process control system, such as in the prior art.

Fig. 2 illustrates the basic concept of the invention.

Fig. 3 illustrates a preferred embodiment of the invention involving a successive approximation algorithm.

Fig. 4 illustrates a second preferred embodiment involving a ramp generator.

Fig. 5 illustrates the way of generating "digital hold" and "analog hold" control signals.

Fig. 6 illustrates a conversion system of the invention allowing simultaneous A/D-amplification and D/A-attenuation conversions.

Fig. 7A and 7B illustrate the basic principle of the use of the invention in an echo cancellation architecture.

Fig. 8 illustrates with details the use of the invention in an echo cancellation architecture in a standard 4-wire modem.

Fig. 9 is an example of an accurate subtracter needed in preferred embodiment of the invention.

In figure 2, the basic elements making up the an embodiment of the invention, referenced as block 100, are described. Block 100 includes a digital to analog (D/A) converter 110, the output of which is connected to an attenuating device 120 by means of lead 135. The output of attenuator 120 is connected to a first input of a comparator 150 and to a sample/hold (S/H) circuit 130 by means of lead 145. The output of S/H circuit 130 provides on lead 185, the analog value corresponding to the digital output transmitted to control logic 140 on bus 115. In order to perform the analog to digital conversion, block 100 also includes a second sample/hold circuit 160 receiving the analog value to be converted on lead 165 and having its output connected to a second input of comparator 150 by means of lead 155. The output of comparator 150 is transmitted to Control Logic 140 by means of lead 175. Control Logic respectively controls S/H circuit 160, D/A converter 110, attenuator 120 and S/H circuit 130 by means of lead 117, busses 127 and 137 and lead 147. Control Logic 140 eventually provides on bus 125, the digital equivalent of the analog value entered in S/H circuit 160.

The digital to analog conversion associated with an attenuation step operates as in the following: The digital information to be converted enters Control Logic 140 by means of bus 115. It should be noticed that the transmission of the digital value can be performed as well serially as in parallel, such as in figure 2. Control Logic 140 transfers on bus 127 this digital value to D/A circuit 110 which provides the analog representation of the digital value to the programmable attenuator 120. The attenuated analog value thus calculated is memorized into S/H circuit 130 in order to release previous blocks for another conversion, be it analog to digital or the inverse, digital to analog. It should be noticed that the D/A can be one of various types well known in the background art: for instance, it may be a flash D/A converter using resistor networks, current sources, capacitor networks or any mix of these. The D/A circuit can also be made up by means of a ramping D/A converter using a current source charging a capacitor, or even several ramps of different values at the same time or cascaded. Programmable attenuator 120 may also be one of well-known attenuating device in the background art. It is generally made of a resistor network with cascaded cells and analog switches controlled by Control Logic 140. However, it can also be made of capacitors and switches synthesizing the resistors using switched capacitor techniques widely used with CMOS technology. Both S/H circuits 130 and 160 generally include an operational amplifier, a holding capacitor and means for inhibiting the input stages of the operational amplifier. This technique is well suited for bipolar technology. However, CMOS technology generally involves some different techniques, and particularly an operational amplifier, at least one integrated capacitor and two CMOS switches.

The analog to digital conversion associated with an amplification step is performed as in the following:

The analog signal existing on lead 165 is first entered into S/H circuit 160 in which, a sample is memorized. As mentioned previously, the advantage provided by S/H circuit 160 is that the analog-to-digital conversion can be started, interrupted as soon as an analog sample is held into S/H circuit 160, then followed by a digital-to-analog conversion step and eventually resumed in the state it was when the interruption occurred. However, if the possibility of a interruption is not required in the analog-to-digital process, then S/H circuit 160 may easily be suppressed. In order to perform the analog-to-digital conversion followed by an amplification step, Control Logic 140 generates a sequence of successive digital values on bus 127. These are converted into their analog representation by D/A circuit 110 and then attenuated by means of attenuator 120. The output of attenuator 120 is transmitted to be compared in comparator 150 with the analog value previously sampled and hold in S/H circuit 160. Comparator 150 is used to determined the sign of the difference between its two inputs so that Control Logic 140 may determine the best digital approximation of the amplified analog input value.

Assuming that k is the value of the attenuation performed by attenuator 120 ( 0< k < 1). At the end of the D/A conversion, one may write the following relation:



(Digital Value provided by CL 140) x k = analog value loaded into S/H 160



therefore:



Digital value = (1/k) x (analog value loaded into S/H 160)

Since the value k is comprised between 0 and 1, 1/k may vary from 1 to infinity. Thus, an analog-to-digital conversion associated with an amplification step, with the transfer function of the cascade A/D-amplification being accurately inverse to that of the cascade D/A-attenuation, has been performed.

The analog-to-digital conversion associated with an amplification step is then completed. Control Logic 140 transfers to bus 125, the above digital value which is the already mentioned best digital approximation of the amplified analog value.

The generation of the sequence of successive digital values performed by Control Logic 140 can be achieved by means of different algorithms. For instance, the above generation may be made on the basis of a successive approximation. In this case, Control Logic 140 first generates a digital value having the most significant bit (MSB) set to one and all remaining bits set to zero. If the output of attenuator 120 is lower than the analog value loaded into S/H circuit 160, this comparison being performed by comparator 150, Control Logic 140 keeps the MSB set and resets it otherwise. Then, Control Logic 140 sets the second bit to one and the second digital value so produced, is processed similarly. This algorithm has the advantage to be generally the fastest of all, as long as the A/D converter embodies a D/A converter.

Another algorithm may be used for generating the sequence of digital values. This one involves the generation of ramps sweeping all possible digital values. This method is much longer than the former one, especially when a great number of bits is involved in the digital value. However, it has the advantage of providing inherently monotonous conversions.

With respect to figure 3, a preferred embodiment of the invention is illustrated. In this embodiment, the generation of digital values performed by Control Logic 140, is made on the basis of a successive approximation. In addition to the already involved blocks and mentioned with respect to figure 2, the invention now involves two registers 210 and 220. Register 210 is an usual 8-bits register designed for storing the value of the digital word to be processed and transmitted by bus 115. For simplicity purpose, figure 3 illustrates an embodiment of the invention involving 8-bits digital words, but words of more or less than 8-bits could be used in the same way. Register 220 stores the value of the digital word provided by the A/D-amplification process. The output of register 210 is connected to a series of 8 NAND gates 231 to 238. Only two numeral references 231 and 238 have been indicated in figure 3 for clearness purpose. The most significant bit (MSB) of register 210 is connected to a first input of NAND gate 231. The second most significant bit of register 210 is connected to a first input of NAND gate 232, and so on. Therefore, the less significant bit (LSB) of register 210 is connected to a first input of NAND gate 238. Every second input of all NAND gate 231 to 238 is connected to "A/D to D/A" control lead 271 which is also connected to the input of an inverter 230.

The outputs of this first series of 8-NAND-gates 231 to 238 are connected to a second series of 8-NAND-gates 221 to 228. For clearness purpose, only two numeral references have again been indicated: 221 and 228. The connection of the first series of 8-NAND-gates to the second series is made according to the following: the output of the first NAND gate 231 of the first series is connected to a first input of the first NAND gate 221 of the second series. Similarly, the output of the second NAND gate 232 of the first series is connected to a first input of the second NAND gate of the second series, and so on. All outputs the second series of NAND gates 221 to 228 constitute an 8-bit-bus which is transmitted to D/A converter 110 and also to the input of register 220 in order to get out the digital result of the A/D-amplification processing after each conversion cycle. All second inputs of the second series of NAND gates 221 to 228 are connected to the outputs of a third series of 8-NAND-gates 211 to 218. Only the numeral references corresponding to NAND gates 211 and 218 have been indicated in the figure. The connection between both second and third series of NAND gates are as in the following: the output of the first NAND gate 211 of the third series is connected to the second input of the first NAND gate 221 of the second series. Similarly, the output of the second NAND gate 212 of the third series is connected to the second input of the second NAND gate 222 of the second series, and so on. At last, the output of NAND gate 218 of the third series is connected to the second input of the NAND gate 228 of the second series. All first inputs of the third series of NAND gates 211 to 228 are constituting an 8-bit bus that receives an 8-bit word from a SAR block 276.

The digital-to-analog conversion is performed as in the the following. Attenuator 130 is adjusted to the required attenuation value by means of a "Att/gain Ctr" control bus 263. "A/D or D/A" lead 271 is set to a high level corresponding to a Digital-to-analog conversion. This entails the locking of all NAND gates 211 to 218 of the third series of NAND gates by means of inverter 230, the output of which being set to a low level. The digital word to be processed is then entered and stored into register 210 by setting to high level "Digital in hold" lead 264. It should be noticed that register 210 is necessary only when the digital word to be converted is not guaranteed to remain stable during one D/A conversion cycle. The digital word entered is therefore converted by D/A converter 110, then attenuated by attenuator 120 and then transmitted to the input of S/H circuit 130 on lead 145. "Analog out hold" control lead 267 is set to a high level so that the analog value loaded into the latter circuit 130 is kept at the analog output lead 185.

The analog-to-digital conversion followed by an amplification step is performed as in the following:

Firstly, programmable attenuator is controlled to provide the required attenuation step by means of "Att/gain Cntrl" lead 263 already used previously. If the value carried on the latter lead remains unchanged, the analog value entered into S/H circuit 160 will be processed with a transfer function being exactly the inverse of that that processed the digital word entered into register 210 previously. Thus, the overall gain provided by the analog-to-digital processing is opposite, from a decibel standpoint, to the attenuation provided by the digital-to-analog processing. The analog value existing on lead 165 is sampled and held in S/H circuit 160 by means of a high level on "Analog in hold" lead 265. "A/D or D/A" lead 271 is set to a low level, which locks the first series of NAND gates 231 to 238 and unlocks the third series of NAND gates 211 to 218 by means of inverter 230.

An "Interrupt A/D" lead is then set to a high level so that a clock signal existing on lead 261 is transmitted through a NAND gate 240 to a successive approximation register (SAR) 276. The latter register 276 is a circuit for providing digital words according to the successive approximation algorithm described above. This circuit may be any of the circuit currently available in the commerce, usually providing 8, 12 or 16 bit-words. In order to start the analog-to-digital conversion, "start A/D" control lead 250 is set to a high level and reset at the same clock period. Consequently, SAR circuit 276 produces a first digital word having its MSB set to one and the other bits set to zero. According to the result of the comparison performed by comparator 150 and transmitted to SAR circuit 276 by lead 175, SAR circuit amends or not the MSB and produces a second digital word with the amended MSB or not and the second most significant bit set to 1 and bits 3 to 8 set to zero. This second word is processed and the second significant bit is amended, if needed, consequently. At the end of the conversion, "A/D complete" lead 260 is set at a high level by SAR circuit 276. The "Digital out hold" control lead 266 is also reset in order to store the SAR output into register 220. Then, the result of the A/D-amplification processing may be taken from bus 125 by an external device.

Because of NAND gate 240, the A/D conversion may be interrupted by setting "interrupt A/D" lead 240 to a low level. This entails the vanishing of the clock signal existing on lead 261 at the input of SAR circuit 276. Consequently, the switching of "A/D or D/A" lead will allow a prioritary digital-to-analog conversion to to be performed. Switching back the latter lead and reactivating the clock at the input of SAR circuit 276 will allow analog-to-digital conversion to be resumed where it was interrupted.

The preferred embodiment so described involved wired logic. The advantage of doing so contrary to the use of a processor implementation comes from the speed of the digital-to-analog conversion thus allowed. However, it should be noticed that the man of the art can use the same principle in order to design a processor implementation involving less components than its equivalent wired logic solution.

With respect to figure 4, a second preferred embodiment is described, in which the analog-to-digital conversion is achieved on the basis of a ramp generation. This embodiment has the disadvantage to be slower than that described below. It may involve either processor or either wired logic technology. For clarity's sake, it illustrates an wired-logic embodiment using a single ramp. This embodiment includes, in addition to the elements already mentioned with respect to figure 2, an Up/down counter 310 for producing a sequence of successive digital words in parallel with the generation of a analog ramp produced by a ramp generator 340. Up/down counter 310 may be loaded with the digital word to be converted by means of a series of AND gates 301 to 308 which are controlled by a single "data/0" lead 357. The connection between bus 115 carrying the digital word and up/down counter 310 is as in the following: the most significant bit of the digital word to be converted is connected to a first input of the first AND gate 301. Similarly, the second most significant bit of bus 115 is connected to a first input of the second AND gate 302 of the series of AND gate 301 to 308, and so on. Every second input of all AND gates of this series is connected to a "data/0" lead 357.

The digital-to-analog processing is initiated by the following steps: "Reset/enable" lead 358 is set to a high level, which resets the ramp generator 340. Similarly, "data/0" lead 357 is set in order to allow the digital word to be transmitted through the series of AND gates 301 to 308 to up/down counter 310. The loading of this digital word is performed by a high level on "load" lead 353. Also, "up/down" lead is set to a level corresponding to a decrementing operation of up/down counter 310. After having adjusted the proper attenuation of attenuator 120 controlled by means of "Att/gain control" bus 263, "reset/enable" lead 358 is switched so as to allow the simultaneous start of both ramp generator 340 and up/down counter 310. This is achieved by means of an inverter 360 receiving the level existing on "reset/enable" lead 358 and controlling up/down counter 310. From this instant, up/down counter starts counting from the digital value loaded by means of the series of AND gate 301 to 308 and down to zero. The output of up/down counter 310 is an 8-bit bus driving a register 320 and decoded by an 8-bit NOR gate 350. As soon as up/down counter 310 reaches the digital value zero, the latter is decoded by NOR gate 350 which raises the level of "D/A complete" lead 354. Consequently to the switching of "D/A complete" lead 354, "analog hold" lead 356 is set in order to store into the first stage of a twin S/H circuit 330 the analog value produced by attenuator 120. At the next digital-to-analog processing, twin S/H circuit 330 transfers the analog signal from its first stage to its second stage which is eventually presented to "analog out" lead 185. Twin S/H circuit is necessary when the conversion time is a significant part of the conversion cycle. In the reverse case, an usual S/H circuit such as that used with respect to figure 3, may be used. A feedback loop is inserted into the D/A processing by means of a lead 362 which connects the MSB at the output of Up/down register 310 to ramp generator 340. Basically, this MSB is integrated to generate a MSB average signal. The latter is used by ramp generator 340 to adjust the offset of its output so that the middle of the full analog scale, generally the middle of the power supply voltage, coincides with the output of ramp generator 340 at the instant when MSB switches.

The analog-to-digital processing is achieved as in the following: "Data/0" lead 357 is reset in order to reset ramp generator 340 by means of "Reset/enable" lead 358.

Subsequently, up/down counter 310 is preset to the digital value zero by "data/0" lead 357. "load" lead 353 is set and "up/down" lead 352 is set to a level corresponding to an incrementing operation of up/down counter 310. The attenuation of the analog output of ramp generator 340 provided by attenuator 120 is adjusted to a proper value by means of "Att/gain control" bus 263. As mentioned previously, in the case where "Att/gain control" bus remains unchanged, the overall gain provided by the A/D-amplification process will by exactly inverse of the attenuation provided by D/A-attenuation process. Then, "reset/enable" lead 358 is switched so that both ramp generator 340 and up/down counter 310 start simultaneously. When the output of attenuator 120 reaches the value of analog value loaded by lead 155, comparator sets "A/D complete" lead 359. Consequently, "digital hold" lead 355 is set and the counter output is transferred into register 320 and the digital value is made available at "digital out" bus 125.

With respect to figure 5, there is described how signals existing on "digital hold" lead 355 and on "Analog hold" lead 356 can be generated. The circuit shown involves four NAND gates 460, 470, 480 and 490 making up two latches. Signal on "A/D complete" lead provided by comparator 150 is transmitted through an inverter 410 to the first input of NAND gate 460. The second input of NAND gate 460 is connected to the output of NAND gate 470. Conversely, the output of NAND gate 460 is connected to a first input of NAND gate 470. Reset signal on lead 358 is transmitted to a first input of a NAND gate 440, the output of which is connected to the second input of NAND gate 470. The second input of NAND gate 440 is receiving "Up/down" signal on lead 352 which is also transmitted through an inverter 430 to a first input of a NAND gate 450. Its second input receives "Reset" signal from lead 358. The output of the latter NAND gate 450 is connected to the first input of NAND gate 480.

The second input of NAND gate 480 is connected to the output of NAND gate 490. Conversely, the output of NAND gate 480 is connected to a first input of NAND gate 490. The second input of this gate receives the complement signal of "D/A complete" after it has been processed by an inverter 420.

For explanation purpose, A/D and D/A processing have been implemented as exclusive. Actually, they can be performed simultaneously on the condition that they are synchronous. In this particular case, two separate counters are needed: one for the counting up and the second for counting down. This is illustrated in figure 6: an "up" counter 510 and "down" counter 520 can operate simultaneously. The digital-to-analog processing is initiated by the following steps: "Reset/enable" lead 358 is set to a high level, which resets the ramp generator 340. In contrary with what is preceding, the digital word to be converted is entered directly into "down " counter 510 by means of "load" lead 352. After the adjustment of the proper attenuation value of attenuator 120 controlled by means of "Att/gain control" bus 263, "reset/enable" lead 358 is switched so that to allow the simultaneous start of both ramp generator 340 and "down" counter 520. This is achieved again by means of inverter 360 receiving the level existing on "reset/enable" lead 358 and controlling both "down" counter 520 and "up" counter 510. From this instant, "down" counter 520 starts counting from the digital value loaded previously down to zero at the speed of clock signal existing on lead 351. As soon as "down" counter 520 reaches the digital value zero, the latter is decoded by NOR gate 350 which raises the level of "D/A complete" lead 354. Consequently, "analog hold" lead 356 is set, according to the part of the description relating to figure 4, in order to store into the first stage of twin S/H circuit 330 the analog value produced by attenuator 120. Similarly as above, at the next digital-to-analog processing, twin S/H circuit 330 transfers the analog signal from its first stage to its second stage which is eventually presented to "analog out" lead 185.

The analog-to-digital processing is achieved simultaneously to the operations above: Ramp generator 340 is reset by means of "Reset/enable" lead 358. Lead 352 is set, which initiates the loading into "up" counter 510 of the digital word zero always existing on bus 515. Then, "reset/enable" lead 358 is switched so that both ramp generator 340 and "up" counter 510 (and also "down" counter 520) start simultaneously. When the output of attenuator 120 reaches the value of analog value loaded by lead 155, comparator sets "A/D complete" lead 359. Consequently, "digital hold" lead 355 is set and the counter output is transferred into register 320 and the digital value is made available at "digital out" bus 125. The generation of "digital hold" and "analog hold" signals on lead 355 and 356 is achieved similarly as above with respect to figure 5. As a conclusion, the simultaneous processing of both A/D and D/A conversion may be summarized as follows: when "Reset/enable" lead 358 is set, ramp generator 340, "up" counter 510 and "down" counter 520 are reset. When "load" lead 352 is set, "up" and "down" counters 510 and 520 are respectively loaded with the digital zero and the digital word to be converted. At the switching of "Reset/enable" lead 358, ramp generator 340, "up" and "down" counters start simultaneously. When "up" counter reaches a value entailing the switching of comparator 150, this digital value being the result of the A/D-processing is loaded into register 320. Also, when "down" counter reaches the value zero, decoded by NOR gate 350, the corresponding analog value at the output of attenuator 120 is stored into Twin S/H circuit 330. Since both A/D and D/A processing use the same components, their transfer functions are exactly inverse each other.

Another possibility to provide both A/D and D/A processing simultaneously would be to use a single counter and a coincidence circuit for comparing the counter output to the digital input to be converted. The counter always counts up and is always loaded with the digital value zero.

Figures 7A and 7B illustrate a typical use of the invention in a echo cancellation device. Indeed, high speed full-duplex data communication on a single channel is of immense practical interest since it involves simultaneous transmission and reception over the same line. A technique for achieving this goal is to provide a mechanism to ensure that the transmitted signal is not fed back into the receiving section of the same end of the line. A transmitter 610 and a receiver 620 are jointly coupled to a two-wire line 640 via an hybrid 630. In an environment of changing channel characteristics (e.g., switched network), the hybrid balancing , if fixed will at best provide a compromise match to the channel. In this mode, a vestige of the local transmitted signal, leaking through the hybrid, can be expected to interfere with the incoming signal from the far-end simultaneously operating transmitter. Figure 7A illustrates the system without any echo cancellation technique and figure 7B illustrates the system using such a technique. As shown in figure 7A, signal entering into receiver 620 is



R + e

The first term represents the signal from the far end and the second term is the echo signal coming from the mismatch of hybrid 630 and the channel 640. Decisions are made by quantizing samples of receiver's output. A typically encountered echo component arising in a system with a conventional compromise balanced hybrid will cause an unacceptably high error rate. To remove the interfering echo component, the local receiver must perform echo cancellation as in figure 7B; that is, estimate by means of an echo estimator 650 the echo signal ê and subtract it in block 660 from the incoming j signal existing on lead 615 prior to making decisions. The estimation is performed by processing the transmitted signal T on lead 605, the received signal R + e on lead 615. This goal is generally achieved by transversally filtering of the local data symbols b(n). If the b(n) are binary, the implementation is simple, requiring mainly additions and subtractions.

However, since the level of the received signal is likely to vary significantly, the far-end signal R from the remote modem may vary between 0 and -43 dBm, it is essential that proper scaling be done when the error signal is computed. Such scaling involves a gain adjust device, or AGC between the analog to digital conversion prior to echo estimation computing. The result of this computation provides the digital value of estimated echo ê which must be converted and then attenuated before being subtracted to the received signal. Figure 8 illustrates such a device: the received signal from hybrid 630 is entered in AGC 750 for proper scaling prior to any computation so as to make the best use of an analog-to-digital converter 730. After amplification and after conversion into digital form of the received signal R + e by A/D block 730, the latter is then entered into echo evaluator 710 which extracts the estimated echo. This is achieved by comparing the sequence of digital values b(n) on lead 605 to the output of A/D 730. The result of this comparison is used in the adaptive process of the adjustment of a digital filter designed to generate the estimated echo. Generally, the tap coefficients of this digital filter are chosen to minimize, in a mean-square sense, the measured receiver error signal which is the difference between the actual receiver output (R+e) on lead 615 and the ideal output built from b(n). Obviously, any other way of choosing the tap coefficients may be used.

Once extracted, the estimated echo ê is transmitted to D/A block 720 in order to be converted back to analog form. It is then attenuated by attenuator 740 before being subtracted in subtracter 660 to the actual received signal (R + e) after it has been delayed by block 780. This delay is inserted so as to compensate the processing delay between signal on wire 615 and signal on wire 625. The gain provided by AGC 750 and the attenuation produced by attenuator 740 are controlled by echo evaluator 510 by means of control leads 770. As shown in figure 8, transmitter 610 and receiver 620 may form the frame of a standard modem 760 having no echo cancellation in itself. Therefore, the use of the invention in this product may add, as an extra feature, such a capability only existing, before, in highly sophisticated and expensive data control equipments.

As mentioned previously, the transfer functions of A/D-amplification block and D/A-attenuation block must be accurately inverse each other. Indeed, it should be noticed that the residual echo e can be some 30 dB above the far-end signal R. Since good performance at high speeds imply a S/N ratio in the 30 dB, it can be seen that the echo must be regenerated with an error less than 60 dB below the echo itself. Therefore, the interest of the architecture of the present invention becomes obvious: since both A/D-amplification and D/A-attenuation blocks are consisting of the same elements, their characteristics are exactly identical. The subtraction performed in subtracter 660 must also have a precision tolerance of about 0.001 in order to allow a good efficiency of the echo cancellation device. A subtracter having such a tolerance may easily be manufactured by means of an operational amplifier 800, such as in figure 9, connected so as to build a differential amplifier by means of 4 resistors R1, R2, R3 and R4. The values of those resistors can easily be chosen to be accurately equal by laser trimming on the same substrate.

It should also be noticed that the amplification block 750 is associated with A/D block 730 since most often, the analog signal to be processed by digital processing means, such as echo evaluator 710, must be amplified for proper scaling before A/D conversion. However, except this point, the association of a A/D conversion and an amplifier is arbitrary and the invention could obviously be applied to provide a system for performing either a A/D-attenuation and either a D/A-amplification step.