FR2643523A1 - Discriminator for digital transmissions - Google Patents

Abstract

The discriminator is intended to be connected to a transmission medium 1 carrying, in accordance with a chosen multiplexing, information of various types which change over time, such as "data", "frequency", "speech", "silence". For each signal sample, processing means 3 determine a value representative of its energy, and effect a filtering of the adaptive linear prediction type so as to generate an optimal set of prediction coefficients. Analysis means 4 can discriminate the type of each information item through a chosen mechanism for decisions resting on its energy and its optimal set of prediction coefficients.

Description

Discriminator for digital transmissions
The invention relates to the study and monitoring of equipment for the digital transmission of information.

Such equipment is often called digital paths when they are intended to transmit digital speech signals and digital data signals interchangeably with, of course, "rests", and, in addition, digital signaling information which does not are of no interest to the end user.

Work has been carried out to distinguish silence from speech, with the aim of transmitting only the "useful" signal. Their purpose is therefore to allow compression of the speech signals, that is to say to reduce the amount of information to be transmitted in order to correctly reflect such signals. This kind of technique is applied in today's concentrators which operate, for example, at the rate of 2,048 Mbits / sec.

However, it currently appears necessary to also make equipment for monitoring "digital paths" of transmission.

Their purpose is quite different, since it is not a question of intervening to compress the speech signals, but rather of producing statistics on the respective shares of occupation of the speech, data and "inactivity" signals in digital paths at 2,048 Mbits / sec.

The term “data” is understood here to mean the signals coming from modems, that is to say analog signals digitized in a frequency band between 300 Hz and 3400 Hz.

Current proposals are based on the energy properties of signals. They have resulted in limited devices as to the number of channels they can process, as well as with regard to their accuracy.

The main object of the invention is then to propose a traffic analysis device which is precise and efficient as regards the number of discriminated signals. Since "silence" cannot be considered as a digital signal proper and must also be discriminated against, the term "information" will be used in the rest of the text for any signal or silence.

The invention therefore relates to a discriminator for digital transmission, intended to be connected to a transmission medium forming a "digital conduit" series and defining, according to a chosen multiplexing, a plurality of digital channels each capable of conveying information of various and evolving types in time.

According to a general characteristic of the invention, this discriminator comprises - pre-processing means, connected to the digital path, and capable of distinguishing the digital channels, taking into account multiplexing, - processing means comprising
energy calculation means capable of providing
for each channel a representative value
of the information energy it carries,
over a chosen time period and,
predictive type filtering means
adaptive linear, able to determine, for
each channel, an optimal set of coefficients
prediction, minimizing the gap between said
information on said channel and its prediction, and - means of analysis capable of discriminating the type of each information by a chosen mechanism of decisions relating to its energy and its optimal set of prediction coefficients.

The invention starts from the observation that the use of filtering of the adaptive linear prediction type makes it possible, with suitable arrangements, to provide a solution to the problem posed.

Other advantages and characteristics of the invention will appear on reading the detailed description below and the appended drawings, in which - FIG. 1 is a schematic block diagram of an embodiment of a discriminator according to the invention - Figure 2 is a schematic block diagram of part of the discriminator processing means of Figure 1; - Figure 3 is a very schematic representation of an embodiment of another part of the means for processing the discriminator of Figure 1 - Figure 4 is a schematic diagram of the means of analysis of the discriminator of Figure 1; FIG. 5 illustrates ranges of prediction coefficients valid for a filter of order 2; and - Figures 6 and 7 illustrate a flowchart of operation of the discriminator according to the invention.

The drawings essentially contain elements of a certain character. As such, they form an integral part of the description and can not only serve to better understand the detailed description below but also contribute, if necessary, to the definition of the invention.

As illustrated in FIG. 1, the discriminator is advantageously connected to a serial digital conduit 1 via a high impedance probe 20 thus making it possible to derive the information to be analyzed without disturbing the digital conduit.

In the application described, this digital path has two directions of transmission 10, 11 (transmission and reception) and conveys digital information according to a known multiplexing of the MIC (pulse code modulation) type.

According to a known standard, for which the nominal bit rate is 2.048 Mbits / sec, the transmission is done on 30 digital channels. It uses a compression / expansion coding law, which does not intervene directly in the present invention. The same standard specifies the structure of the multiplexing used in this transmission, which is important to remember.

This multiplexing is subdivided in the following way - it is firstly decomposable into a series of multiframes, each of a duration of 2 ms, note being made that the discriminator does not use here the cutting proper in multi-frames for its treatments ; - each multi-frame is in turn broken down into 16 frames, each of which lasts 125 microseconds - each frame is finally subdivided into 32 words, each of duration 3.9 microseconds. The words of ranks 1 to 15 and 17 to 31 define the 30 channels of effective digital transmission; the words of ranks 0 and 16 are used for the transmission of digital service information.

The digital information conveyed on the pipe 1 is of the type with built-in clock coded in the form known as HDB3. Each piece of information is assigned to a channel and is therefore sampled at 8 kHz, each sample consisting of the word in the frame corresponding to the channel. In this application, the words are bytes each having 8 bits.

A frame can contain information of various types and the same channel can convey, following each other, two information of different types. The discriminator according to the invention makes it possible to distinguish the following types of information - silence, or inactivity, - speech, - bit rate data less than or equal to 4800 bits / sec, - bit rate data greater than 4800 bits / sec, - frequency conditioning at 2100 Hz, generally preceding the data information, - pure frequencies belonging for example to the CCITT signaling codes 4 and 5.

It is recalled here that the information of the given type is contained in the frequency band 300-3400 Hz.

The two outputs (transmission and reception) of the high impedance probe 20 are respectively connected to two identical HDB3 information decoders 21. Each of these decoders 21 has a clock output making it possible to restore the time base used in the digital path as well as an information output by which it delivers, from the words of the frame, parallel words formed each of the bits of a sample of a digital channel. The issuance of the clock signal thus makes it possible to provide an indication of the rank of the parallel word, designating the channel to which it belongs.

The outputs of the two decoders 21, which correspond to the outputs of the pre-processing means 2, are connected to processing means 3. The latter are identical for the two directions of transmission and they will only be described in detail for one of them.

The clock and information outputs of the decoder 21 are firstly connected to a first processing block 30 comprising (FIG. 2) means of expansion 300 of the parallel words coming from the decoder 21. These means of expansion 300 provide, from an 8-bit "compressed" parallel word, an expanded parallel word whose information content is coded in 13 bits, including the sign bit. We will then say that such a sample has gone from a compression coded form of approximate logarithmic type (8 bits) to a linear form (13 bits). However, the 13-bit sample can be represented using a 16-bit word using the 2's complement and repeating the sign bit three times.

The 30 16-bit samples (30 transmission channels) are then distributed at the output of the expansion means 300 between six microprocessors for signal processing 31-1 to 31-6 (FIG. 3) connected in parallel on a serial bus 16-bit common B. The other six microprocessors assigned to the 30 reception channels are also connected on this same common bus B and the twelve microprocessors thus form filtering means 31, of the adaptive linear prediction type as will be seen below. after. Each microprocessor 31-i is for example that marketed by the company TEXAS INSTRUMENT under the reference
TMS320C10.

The block 30 also includes energy calculation means 301,303 able to supply for each channel, a value representative of the energy of the information that it conveys. This energy is calculated over a chosen time period T equal here to the sampling period, ie 125 microseconds. Initially, these energy calculation means 301 provide a "logarithmic" energy value from the 8-bit parallel words. The principle of calculation of this energy is as follows: for a given channel, at each instant, the amplitude of the information sample is compared to the effective value of the energy calculated from the previous sample of this same channel t if the amplitude is higher, the effective value is incremented by a fixed step, if not, it is decremented by the same fixed step. For more details, those skilled in the art will refer to the article.
An efficient step size adaptation Technique for LMS adaptive filters, Maurice BELLANGER and C. Cengiz EVCI, pages 1153-1156, 1985 IEEE.

From this first (logarithmic) energy value, the calculation means 303 determine, using a bijective correspondence table, a second so-called "linear" energy value corresponding to the energy of the expanded parallel words. .

In addition to these energy calculations, calculation means 304 determine for each sample, the adaptation step of the linear prediction filter which will be used to determine the type. This adaptation step p depends, in general, on the energy of the sample as well as on a parameter of speed of convergence of the filter v, identical for all the channels.

More precisely, the adaptation step p depends on the ratio of the parameter v on the value of the energy. We use the value of the "linear" energy here but we could use the "logarithmic" value of this energy.

It should also be noted that the parameter v is chosen once and for all depending on the nature and the number of types of information to be distinguished. This parameter is advantageously of the order of 2-5.

To distinguish the different types of information described above, the mere use of energy conditions is clearly insufficient. Also, the Applicant has observed that the exploitation of the spectral properties of this information and, in particular, their autocorrelation factor constitutes a good criterion of distinction. Indeed, the pure frequencies are entirely deterministic therefore totally correlated; speech is correlated in the short term while data, which can be likened to noise, is fully decorrelated.

However, the study of the spectral properties of said information must be carried out in a sufficiently simple manner to be able to be compatible with real-time processing and must moreover present an industrially reasonable cost.

Also, the Applicant has surprisingly found that the study of the prediction coefficients, delivered for each sample by an adaptive linear prediction filter, made it possible to achieve the desired goal. Indeed, the general principle of an adaptive linear prediction filtering consists in determining, for each sample, an optimal set of prediction coefficients minimizing the difference between the information sample and its prediction. When this difference is minimal, the value of these coefficients as well as their temporal stability which depends on the autocorrelation of the sample, constitutes in a way a signature of the information and helps to distinguish the type.

The filtering means 31 compare each sample x with the corresponding predicted sample so as to provide an error signal e. Those skilled in the art know that each prediction coefficient ai is given by the following formula
ai (t + l) = ai (t) + pe (t) .x (ti) in which t denotes the current time and i is less than or equal to Nl, N denoting the number of coefficients of the filter.

In this application, each filtering means 31 has the same number of coefficients advantageously taken greater than or equal to two and less than 4. The purpose of the adaptation is to provide an optimal coefficient set minimizing the error signal e.

The various coefficients are sent over 15 bits via two 16-bit serial buses (one for transmission, one for reception), to analysis means 4 which include a single fast signal processing microprocessor such as the one with the reference
TMS320C10.

The sixteenth bit of the word sent to the analysis means is representative of the sign of a counter; its meaning and usefulness will be explained in detail below.

We only describe here the development of the value of this sixteenth bit.

For each sample x (t), taken at time t, the filtering means 31 compare the value of its amplitude (denoted Vcrest (t)) with the second value e2 of its energy ("linear" value). If the ratio V peak / e2 is greater than a threshold value, determined during the development of the discriminator and taken here equal to 10, the counter is incremented. If this ratio is less than said threshold value, the counter is decremented. If the linear value e2 of the sample energy is less than a linear reference threshold (meaning as we will see below that the sample is of the "silence" type), the counter is reset to zero. The sixteenth bit of the word sent to the analysis means is then representative of the sign of the value of said counter.

The microprocessor of the analysis means as well as its associated memories are illustrated diagrammatically in FIG. 4.

A reference table 41 contains values of reference coefficients which define areas representative of the different types of information. A particular example for a filter of order 2 in z of type 1 - Al.z-1 - A2.Z-2 is illustrated in figure 5.

The two coefficients Al and A2 of this filter evolve respectively between -2 and 2 and between -1 and 1 inside a triangle. Ten zones Z1 to Z10 are then defined respectively associated with the following types Z1: zone corresponding to information of the speech type;
Z2: zone corresponding to information of the given type
whose flow is less than or equal to 4800 b / s;
Z3: zone corresponding to type information
following: speech, given at a lower rate or
equal to 4800 b / s, given at a flow rate greater than 4800 b / s;
Z4: zone corresponding to type information
following: given at flow less than or equal to
4800 b / s, given at a flow rate greater than 4800 b / s;
Z5: zone corresponding to information of the fre type
quence at 2600 MHz;
Z6: zone corresponding to information of the given type
at a speed less than or equal to 4800 b / s;
Z7: zone corresponding to information of the fre type
quence at 2400 Hz or of the given type at lower flow
or equal to 4800 b / s;
Z8: zone corresponding to information of the fre type
quence at 2100 Hz;
Z9: zone corresponding to information of the fre type
quence at 2040 Hz;
Z10: indecision zone not corresponding to any type
and to which coeffi values belong
cients of a filter that for example has not yet
converged, especially following a change
type of information.

When the table is addressed, it compares the coefficients calculated for a sample with the reference coefficients and returns the reference of the zone in which these coefficients are located.

A calculation unit 42 is on the one hand suitable for comparing, for each sample, the first ("logarithmic") value of its energy to a reference energy threshold here equal to - 40 dB. This threshold is a programmed value depending on the envisaged application. It corresponds to the linear reference threshold described above and used by the filtering means 31 for the development of the value of the sixteenth bit. Of course, the calculation unit could also have used this linear reference threshold to then compare it with the second ("linear") value of the energy of the sample.

This calculation unit addresses the reference table 41 to each set of prediction coefficients, receives in return the reference of the corresponding area and deduces therefrom, using a decision flowchart described below, the type of information. .

Although the analysis means receive the coefficients and the energy values every 125 microseconds, for each channel, the type determination is only carried out for this channel every 15 milliseconds. The calculation unit 42 then compares the decisions, channel by channel, for each of the two directions of transmission and reception and communicates this decision to an interface 5 connected to additional processing means (not shown in the figures), intended to perform statistical operations on the results obtained. It should also be noted that the preprocessing means 2 are able to detect any operating anomaly both on the digital conduit and on the discriminator itself and to transmit these said anomalies directly to the interface means 5.

We will now describe in detail, with particular reference to Figures 6 and 7, the decision mechanism for distinguishing the various types of information.

For each sample, it is determined, in a step 60, if its logarithmic energy is greater than the reference threshold. If not, then the information is considered to be of the silence type.

If so, in a step 61, the prediction coefficients are analyzed by addressing the reference table.

If this analysis leads to the Z10 indecision zone, the corresponding sample is assigned the same type as the previous sample on the same channel.

Otherwise, it is tested (step 62) if, taking into account the reference of the area returned by the table 41, only one type of information is possible.

If so, the discriminated type is confirmed by carrying out, in a step 63, a verification with the past, that is to say that it is verified that a certain number of previous samples of the same channel were of this type. In this application, the number of samples is taken equal to 4, which corresponds to a verification over an elapsed time of 60 ms.

If, on the contrary, in step 62, the reference of the zone shows that there are several possible types of information, we analyze, in step 64, whether the discrimination relates to the speech or data types (for example example zone Z3) or on the frequency or data types (for example zone Z7).

If the discrimination relates to the frequency / data types, in step 65, the temporal stability of the coefficient set is analyzed over a period of 32 samples, ie 480 ms. If the coefficients are considered stable, and the information lasts less than 480 ms, it is then decided that it is a frequency type. Otherwise, the information is assumed to be of the given type.

In the case where in step 64, the frequency type cannot be envisaged, it is determined, using a study of the past (step 66) whether the information in question is of the given or speech type.

In all cases, with regard to the data, the analysis means determine the data rate (step 67). These analyze the modulation of information of the given type. Indeed, the data, whose bit rate is greater than 4800 bits / sec, are both phase and amplitude modulated, while the data whose bit rate is less than or equal to 4800 bits / sec are only modulated phase. The test then consists in an analysis of the sixteenth bit mentioned above, that is to say in a study of the sign of said counter. A positive value of this corresponds to a bit rate greater than 4800 bits / sec, while a negative value corresponds to a bit rate less than or equal to 4800 bits / sec.

The next step, bearing the reference 68, consists of a configuration, channel by channel, of the results of the transmission direction and of the reception direction. If there is an incompatibility between the two types obtained, the most likely is taken according to the priorities established during of the design of the discriminator. The order of priority is in this application (from strongest to weakest): frequency 2100 Hz; frequency other; speech; given
If the two results are consistent, the problem does not arise. Finally, in the latter case, and on the assumption that the information is of the given type, an additional test determines whether the information is preceded or not by information of the frequency type at 2100 Hz.

 Finally, it should be noted, in the case of information of a given type, that the confrontation of the decisions of the two directions of transmission makes it possible to specify whether the transmission of the data is a "ha If duplex" transmission (ie say in one direction) or "full duplex" (that is, in both directions).

Generally, in this application, the additional processing means do not carry out the development of statistical operations during all the periods during which the digital path studied is considered to be in alarm.

The invention is not limited to the embodiments described above but embraces all the variants, in particular the following - the decoder 21 specific to HDB3 information can be easily interchangeable and replaced by other interface means if the bit rate of the digital path changes, in particular if it is equal to 1.544 Mbits / sec, - the invention is not dependent on the type of compression coding used in the digital path; this, carried out in a reprogrammable memory, can be easily interchangeable if the coding used is different, - a type of coding using 8-bit words has been described above; of course, any multiplexing is suitable provided that each sample of digital information is defined by bits contained, in a identifiable manner, in the successive frames of the multiplexing, - the additional processing means, developing the statistical operations, can work simultaneously with by example 8 discriminators of the type described, which provides an analysis capacity of 240 channels.

Claims (19)

Claims 1. - Discriminator for digital transmissions, intended to be connected to a transmission medium (1) forming a "digital conduit" series and defining, according to a chosen multiplexing, a plurality of digital channels each capable of carrying information of various and evolving type in time, characterized in that it comprises - pre-processing means (2), connected to the digital conduit (1), and capable of distinguishing the digital channels, Given the multiplexing, - processing means (3) comprising  energy calculation means (301,303),  able to provide a value for each channel  representative of information energy  that it conveys, over a period of time  chosen, and  filtering means (31) of the predicted type  adaptive linear tion, able to determine,  for each channel, an optimal set of coefficients  prediction, minimizing the gap between said  information on said channel and its prediction, and, - analysis means (4) capable of discriminating the type of each information by a chosen mechanism of decisions relating to its energy and its optimal set of prediction coefficients. 2. - Discriminator according to claim 1, in which each digital information sample is defined by bits contained, in a identifiable manner, in each of the successive frames of the multiplexing, characterized in that the preprocessing means (2) constitute parallel words formed each of the bits of a sample of a digital channel in correspondence with a rank indication designating this channel. 3. - Discriminator according to claim 2, characterized in that the energy calculation means (301) determine, for each channel, a first value el of the energy of the information that it conveys from the comparison between the amplitude of the current sample and the effective value of the energy calculated from the previous sample. of this same way. 4. Discriminator according to one of claims 2 and 3, characterized in that the processing means comprise means of expansion (300) of the parallel words supplied in the pre-processing means, capable of providing expanded parallel words . 5. - Discriminator according to claims 3 and 4, characterized in that the energy calculation means (303) determine for each channel a second energy value e2 corresponding to the first and representative of the energy of the expanded word. 6. - Discriminator according to one of claims 4 and 5, characterized in that the filtering means deliver, for each sample, the corresponding coefficient set by using an adaptation step dependent on the energy of said sample. 7. Discriminator according to claim 6, characterized in that the adaptation step p depends on the ratio v / e2 of a chosen parameter of speed of convergence of the filter v on the second value e2 of the energy of the sample- lon. 8. - Discriminator according to claim 7, characterized in that the chosen convergence speed parameter is the same for all the samples. 9. - Discriminator according to one of claims 6 to 8, characterized in that the number of coefficients is the same for all the samples. 10. - Discriminator according to one of claims 6 to 9, characterized in that the number of coefficients is at least equal to two and less than four. 11. - Discriminator according to one of the preceding claims, characterized in that the filtering means comprise a plurality of processing units (31-i) connected in parallel on a common bus (B). 12. - Discriminator according to one of the preceding claims, characterized in that the analysis means (4) are capable of comparing, for each sample, its value representative of its energy with a reference energy threshold. 13. - Discriminator according to one of the preceding claims, characterized in that the analysis means comprise storage means (41) of a series of reference prediction coefficients, corresponding to the various types of information and means comparison (41) able to compare, for each sample, its set of prediction coefficient with the reference series. 14. - Discriminator according to one of the preceding claims, characterized in that the digital conduit (1) comprises two directions of communication (10,11) and in that the analysis means are capable of comparing (42) for each channel, the two types of the two samples circulating in both directions. 15. - Discriminator according to one of the preceding claims, characterized in that it further comprises interface means (5) with additional processing means capable of carrying out operations of the statistical type. 16. - Discriminator according to claim 15, characterized in that the pre-processing means are capable of detecting any operating anomaly and of transmitting it to said interface means. 17. - Discriminator according to one of the preceding claims, characterized in that the pre-treatment means comprise a high impedance probe (20) connected to the conduit. 18. - Discriminator according to one of the preceding claims, characterized in that the types of information to be discriminated are chosen from the group silence, speech, data, frequency. 19. - Discriminator according to one of the preceding claims, characterized in that the multiplexing is of the pulse code modulation type.