WO1989008360A1 - Procede de detection adaptative pour signaux quantifies - Google Patents

Procede de detection adaptative pour signaux quantifies Download PDF

Info

Publication number
WO1989008360A1
WO1989008360A1 PCT/FI1989/000037 FI8900037W WO8908360A1 WO 1989008360 A1 WO1989008360 A1 WO 1989008360A1 FI 8900037 W FI8900037 W FI 8900037W WO 8908360 A1 WO8908360 A1 WO 8908360A1
Authority
WO
WIPO (PCT)
Prior art keywords
signal
reference numbers
sample value
values
states
Prior art date
Application number
PCT/FI1989/000037
Other languages
English (en)
Inventor
Teuvo Kohonen
Original Assignee
Oy Nokia Ab
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Oy Nokia Ab filed Critical Oy Nokia Ab
Priority to DE19893990156 priority Critical patent/DE3990156T1/de
Publication of WO1989008360A1 publication Critical patent/WO1989008360A1/fr
Priority to GB9018361A priority patent/GB2233865B/en

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/38—Synchronous or start-stop systems, e.g. for Baudot code
    • H04L25/40—Transmitting circuits; Receiving circuits
    • H04L25/49—Transmitting circuits; Receiving circuits using code conversion at the transmitter; using predistortion; using insertion of idle bits for obtaining a desired frequency spectrum; using three or more amplitude levels ; Baseband coding techniques specific to data transmission systems
    • H04L25/4917—Transmitting circuits; Receiving circuits using code conversion at the transmitter; using predistortion; using insertion of idle bits for obtaining a desired frequency spectrum; using three or more amplitude levels ; Baseband coding techniques specific to data transmission systems using multilevel codes

Definitions

  • the invention relates to an adaptive detection method for quantized signals having nominal values which at different moments define a finite set of discrete states, in which method the discrete state of a signal at a certain moment is interpreted by comparing the sample value of the signal to a set of reference numbers, on the basis of which the recognition of the signal state takes place.
  • the transmission path in a data transmission system is never ideal, wherefore the transmission always distorts the signal to some extent, which may hamper the detection of the signal at the receiving end.
  • the transmission path may cause even great random variation, variation and deformations in the representations (signal values) of the discrete states of the signal to be transmitted, which causes errors in detection or makes the detection impossible.
  • digital codes are often transmitted in the form of modulation depths corresponding to their binary number values, that is, by quantized analog variables. Thereby the transmission path may cause great and relatively rapid variation in the level of modulated waves.
  • an adaptive receiving method (equalizer) in the transmission.
  • the sensitivity of the detecting method automatically follows up variations in the level of the signal to be transmitted.
  • the object of the invention is to provide a new adaptive detection method for quantized signals, which reduces the effect of variations and deformations caused by the Transmission line on the interpretation of signal states.
  • At least the reference number selected as an recognition result is corrected on the basis of the distance between the sample value and the reference number after each sample value, or each reference number is corrected at regular intervals on the basis of the cluster means of the sample values grouped into clusters according to the recognition results.
  • received signals are sampled and it is determined by a number of samples as low as possible whether there is statistical clustering present in the samples, whereby the means or other averages of such clusters are observed and used as new reference numbers when detecting various quantized states.
  • the reference numbers follow up the levels of the signals rather accurately, thus improving the detection accuracy.
  • the reference values of the different clusters determined as described above may attain incorrect order and be trapped in this order, which causes severe, prolonged disturbances in reception, because the discrete states of the signals are continually misinterpreted.
  • the principle of self-organizing feature maps is applied to the correction of reference numbers.
  • the object of the detection method utilizing self-organizing feature maps is to restore the metric-topological relations of the reference values according to the corresponding relations present in the signal values, also when the order of the reference numbers is temporarily incorrect.
  • a detection method based on averages and another method based on self-organizing feature maps are used alternately. Such a solution enables an optimal compromise between the follow-up accuracy and the ability to restore the correct order.
  • the detection method according to the invention provides increased accuracy of detection and moderates requirements set on the accuracy of the preceding equalizers or even replaces such equalizers.
  • the method becomes even more advantageous when the number of different signal states increases, because even a minor disturbance may thereby cause errors in detection.
  • the method of the invention is rather light to realize and enables efficient correction even at high transmission rates, so that the detector applying the invention can be realized as a signal processor or microcircuit in a more advantageous manner than conventional equalizers.
  • the rate of previous signal processors is not sufficient for processing signals having a very high speed.
  • the method can be realized by digital or analog computation methods.
  • Figure 1 shows a one-dimensional feature map
  • Figure 2 shows one way of realizing the feature map of Figure 1;
  • Figure 3 shows an example of the function of a method utilizing the feature map
  • Figure 4 shows a two-dimensional feature map
  • Figure 5 shows an apparatus applying the method of the invention.
  • the method of the invention is applicable with a one-dimensional signal (such as a multi-level signal), the values of which are quantized over a domain [a, a + nq], where a is the lowest value of the domain, n is an integer, and q is the quantum step.
  • a is the lowest value of the domain
  • n is an integer
  • q is the quantum step.
  • the aim is thus to give the signals one of the values a + hq, where h is the sequential index of the signal state, 0, 1, ...,n; however, the transmission path alters it into a random value a + hq + 6 h , where ⁇ h is a random error (possibly larger than the quantum step q).
  • the signal states of the received signal may vary within wide limits due to various disturbances; however, within a short time span, the metric-topological relations between the signals remain unchanged.
  • the principle in the present method is to automatically observe by means of the lowest possible number of samples of received signals whether there is statistical grouping (clustering) present in the samples and to produce adaptive reference numbers or vectors best corresponding to these clusters, whereby the reference numbers or vectors can be used for the recognition of quantized signal states.
  • clustering statistical grouping
  • This can be achieved by means of conventional clustering methods or vector algorithms. These include the k-means clustering applied in digital signal technology and pattern recognition, disclosed in the reference: [1] J. Makhoul, S. Roucos, H. Gish, "Vector
  • a set of consecutive signal samples is collected such that each possible signal state occurs in the samples several times.
  • the samples are classified on the basis of the recognition result described under the preceding heading, and the mean of each class is determined.
  • the means taken from the samples or, for instance, a weighed linear combination of the old reference numbers and the means taken from the samples can now be used as new reference numbers m i .
  • the reference numbers follow up variations in the signal values.
  • the updating of the numbers is here performed at regular intervals.
  • the updating of the reference numbers may also be performed after each signal sample, whereby an experimental recognition of the sample according to Eq. (1) has to be carried out first so as to find the closest reference value m c . This and this only is updated:
  • the parameter ⁇ (0 ⁇ ⁇ ⁇ 1) is a coefficient determining the relative magnitude of the corrections.
  • the corrected reference number is obtained in such a manner that the difference between the signal sample x and the original reference number is multiplied with ⁇ and this is added to the original reference number.
  • the preferred embodiment of the method of the invention utilizes so called self-organizing feature maps having this property.
  • FIG. 1 shows a set of processing units called nodes hereinbelow.
  • the lines between the nodes denote topological neighbourhood relations, that is, which nodes are topologically closest to each other.
  • the topological order of the nodes means that the reference numbers m, the intended values of which are closest to each other are located at adjacent nodes in the map representation of Figure 1, that is, belong to the topological neighbourhood of each other. At an erroneous state this order, however, may temporarily differ from that.
  • FIG. 1 The topological map representation of Figure 1 can also be illustrated as shown in Figure 2, in which the nodes, that is, the processing units, are connected in parallel to the signal line, and one and the same signal x is connected to the input of each processing unit.
  • Each processing unit comprises an output y i at which it produces an output signal when the signal state of the signal x is closest to the reference number m i of the node in question. If the signal x is a multi-level signal , for instance, the processing units could be, e.g., comparators.
  • the recognition of the signal states of the received signal is carried out according to Eq. (1).
  • the updating of the reference numbers is carried out by the following correction rule (adaptation rule, clustering rule) : both the reference number m c selected as the recognition result and the reference numbers of the nodes neighbouring to the node c corresponding to the reference number m c are corrected towards the value of x. That is:
  • the parameter a is a coefficient which determines the relative magnitude of the corrections.
  • N c is an index set con ⁇ isting of the node c and its topological neighbours. In Figure 1, nodes which are topological neighbours are interconnected by a line.
  • the above correction rule is able to efficiently restore the magnitude order of the reference number m i so that it is equal to that present in the received signal states, whereas a certain bias is left in the values m i , which will be discussed below.
  • the bias left by the above-described correction rule can be for a major part compensated for and the reference numbers will attain more accurate values, which now approximate the cluster means.
  • the probability density p(x) of the signal x is an arbitrary, but known function, whereby parameters b i and d i compensating for the bias are computable for the precompensation equation.
  • the final weighed correction rule by which the abovementioned bias, too, can be compensated for, is then typically of the form
  • m i (t + 1) m i (t) + ⁇ (b i + d i ⁇ -m i (t)), i ⁇ N c (6)
  • m i (t + 1) m i (t), i ⁇ N c (7)
  • the continuous curves in Figure 3 describe possible signal states, the magnitudes of which vary herein according to a certain arbitrarily chosen law. The actual values of the signal x are picked up at random from these four curves at different moments.
  • the recognition rule can be defined as comparison of distances according to some vectorial norm. Recognition:
  • d( . ) may be some Minkowski distance measure, a special case of which is the Euclidean distance.
  • N' c N' (t) is a neighbourhood set in a twodimensional network of nodes, as shown in Figure 4, for instance.
  • the neighbourhood set N' c of the node c comprises the surrounding nodes closest to the node c, as is shown by the continuous line.
  • the topological order of the nodes can be such, for instance, that the reference numbers m i1 of the reference vectors M i present at the nodes increase from the top towards the bottom while the reference numerals m i2 increase from the left to the right.
  • each reference vector M i correspondingly contains n reference numbers m 1 .
  • the adaptive reference numbers used in the present detection method are solely determined on the basis of the metric-topological relations of the different signal states inherent in the detected signal without any need to identify or label any signal state in the transmitted signal. This method is thus concerned with nonsupervised learning, which is typical of all clustering methods and most adaptive signal processing systems.
  • the present method thus comprises two essential modes: 1.
  • the mode in which the reference numbers of the different signal states are made to follow up the cluster means of the representations of the states. 2.
  • the operations required by the modes 1 and 2 can be implemented numerically, for instance, using so called signal processing architectures constructed of available components, or analog circuits.
  • the choice depends on the required accuracy, computing speed, and costs; the compromise in each particular case is determined by the prevailing state-of-the-art of different circuits technologies.
  • the signal processor solution described in Reference [4] used for speech recognition, is technically possible for the purpose but unnecessarily complicated and expensive.
  • the correction rule can also be implemented by conventional analog computing technology, described, e.g., in G.A. Korn, T.M. Korn, "Electronic Analog and Hybrid Computers", McGraw-Hill, New York, 1964, or equivalent electronic, optic or optoelectronic constructions.
  • the detection in the receiver can be very simple; the demand for accuracy is very modest, because the adaptive control automatically compensates for nonlinearities and instabilities. In fact, it is sufficient for the detector that the magnitude relations of the detected signal states are preserved over a short time span, because the clustering analysis used is able to automa ⁇ ically find the clusters of transmitted signals corresponding to these states and adapt to them.
  • Figure 5 illustrates a typical block scheme of an apparatus applying the method.
  • Part 1 is a means for modulation and transmission of the signals;
  • part 2 is a means for reception and detection (demodulation) of the signals;
  • part 3 is a means in which reference numbers which follow up the received signal states with an accuracy as high as possible (using, e.g., the k-means method);
  • part 4 is a means in which the reference numbers are determined, e.g., by correction rules based on self-organizing map representations, whereby their correct metric-topological relations are restored automatically.
  • Part 6 may receive information from manual control, from the receiver (part 3) of the transmitted signal, which receiver indicates when the transmission is started and interrupted, or from interpretation means (parts 3 and 4) which are able to indicate continual misinterpretation.
  • part 5 may be connected to parts 3 and 4 at regular or otherwise predetermined intervals, controlled by a timer.
  • the initial values of the reference numbers m i or M. contained in parts 3 and 4 may be preset to precomputed values.
  • the control means, part 6 may also set the reference numbers m i or M i to suitable default values.
  • An advantage of the apparatus described is that at the startup of the reception, after interruptions and in case of severe malfunction, the correct order of the reference numbers can be restored by connecting the signal to part 4. At continued reception, the signal can be connected to part 3, which computes the reference values more accurately. In this way it is possible to optimize both accuracy and restoration after disturbances.
  • the adaptive and self-correcting signal detection method of the invention is suitable for use in any kind of data transmission in which the nominal values of the signals to be transmitted at different moments define a finite group of discrete states.
  • the signals may directly convey digital information, e.g. in transmission within a device.
  • the method can be applied when digital signals or states are converted into analog quantities and are transmitted as such, and then again analyzed, detected and coded by this method into digital form.
  • One application is a cellular radio (mobile telephone) in which speech or other analog signals are first digitized by deltamodulation, then converted into analog form (e.g., QAM), transmitted in this form, recoded into digital form and reconstructed into speech.
  • analog form e.g., QAM
  • the present method is not restricted to any particular modulation method or standard in data transmission, but it is generally intended to separate quantized states from each other.
  • the method is also applicable to optic signals or to devices reading digital signals from magnetic or optic memories.

Abstract

La présente invention se rapporte en général à un procédé de détection adaptative qui sert à quantifier des signaux ayant des valeurs nominales définissant à différents moments un ensemble fini d'états discrets, où l'état discret d'un signal à un certain moment est interprété par comparaison de la valeur-échantillon du signal avec un ensemble de nombres de référence, sur la base de laquelle on peut effectuer la reconnaissance de l'état du signal. Selon le principe de la présente invention, les valeurs des nombres de référence utilisés dans la détection sont corrigées de façon adaptative en direction des états de signal effectifs du signal à détecter, par exemple par un procédé de groupage statistique par moyennes k (''k-means'') ou par des procédés de reproduction topologique ayant des caractéristiques d'auto-organisation.
PCT/FI1989/000037 1988-03-04 1989-03-03 Procede de detection adaptative pour signaux quantifies WO1989008360A1 (fr)

Priority Applications (2)

Application Number Priority Date Filing Date Title
DE19893990156 DE3990156T1 (de) 1988-03-04 1989-03-03 Adaptives ermittlungsverfahren fuer quantisierte signale
GB9018361A GB2233865B (en) 1988-03-04 1990-08-20 An adaptive detection method for quantized signals

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
FI881007A FI881007A0 (fi) 1988-03-04 1988-03-04 Foerfarande foer adaptiv avlaesning av kvantiserade signaler.
FI881007 1988-03-04

Publications (1)

Publication Number Publication Date
WO1989008360A1 true WO1989008360A1 (fr) 1989-09-08

Family

ID=8526024

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/FI1989/000037 WO1989008360A1 (fr) 1988-03-04 1989-03-03 Procede de detection adaptative pour signaux quantifies

Country Status (5)

Country Link
JP (1) JP2806584B2 (fr)
AU (1) AU4074389A (fr)
FI (1) FI881007A0 (fr)
GB (1) GB2233865B (fr)
WO (1) WO1989008360A1 (fr)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0465851A2 (fr) * 1990-06-14 1992-01-15 Oy Nokia Ab Récepteur de signaux MAQ qui compense à la fois des distorsions linéaires et non-linéaires
EP0544875A1 (fr) * 1991-06-20 1993-06-09 Universal Data Systems Inc Appareil destine a regler des points de signal, des gains d'egaliseur et autres
EP0574223A1 (fr) * 1992-06-12 1993-12-15 Oy Nokia Ab Procédé de détection adaptive et détecteur pour signaux quantifiés
US5680514A (en) * 1994-09-23 1997-10-21 Hughes Electronics Multiple elastic feature net and method for target deghosting and tracking
WO1999026423A2 (fr) * 1997-11-13 1999-05-27 Industrial Research Limited Prediction de signaux dans des communications mobiles et systemes utilisant ces predictions pour la reception et le decodage de signaux
US10496084B2 (en) 2018-04-06 2019-12-03 Oracle International Corporation Dequantizing low-resolution IoT signals to produce high-accuracy prognostic indicators

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4326169A (en) * 1980-03-07 1982-04-20 Bell Telephone Laboratories, Incorporated Adaptive decision level circuit
US4355402A (en) * 1978-10-19 1982-10-19 Racal-Milgo, Inc. Data modem false equilibrium circuit
US4439863A (en) * 1980-11-28 1984-03-27 Rockwell International Corporation Partial response system with simplified detection
US4528676A (en) * 1982-06-14 1985-07-09 Northern Telecom Limited Echo cancellation circuit using stored, derived error map
EP0180969A2 (fr) * 1984-11-06 1986-05-14 Nec Corporation Circuit de commande automatique de niveau pour un convertisseur analogique-numerique
US4601044A (en) * 1983-11-04 1986-07-15 Racal Data Communications Inc. Carrier-phase adjustment using absolute phase detector
US4602374A (en) * 1984-02-27 1986-07-22 Nippon Telegraph & Telephone Public Corporation Multi-level decision circuit
US4635276A (en) * 1985-07-25 1987-01-06 At&T Bell Laboratories Asynchronous and non-data decision directed equalizer adjustment
EP0238288A1 (fr) * 1986-03-17 1987-09-23 Hewlett-Packard Limited Analyse de transmissions numériques par voie hertzienne

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS61194921A (ja) * 1985-02-22 1986-08-29 Nec Corp 適応ベクトル量子化装置
GB8720387D0 (en) * 1987-08-28 1987-10-07 British Telecomm Matching vectors

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4355402A (en) * 1978-10-19 1982-10-19 Racal-Milgo, Inc. Data modem false equilibrium circuit
US4326169A (en) * 1980-03-07 1982-04-20 Bell Telephone Laboratories, Incorporated Adaptive decision level circuit
US4439863A (en) * 1980-11-28 1984-03-27 Rockwell International Corporation Partial response system with simplified detection
US4528676A (en) * 1982-06-14 1985-07-09 Northern Telecom Limited Echo cancellation circuit using stored, derived error map
US4601044A (en) * 1983-11-04 1986-07-15 Racal Data Communications Inc. Carrier-phase adjustment using absolute phase detector
US4602374A (en) * 1984-02-27 1986-07-22 Nippon Telegraph & Telephone Public Corporation Multi-level decision circuit
EP0180969A2 (fr) * 1984-11-06 1986-05-14 Nec Corporation Circuit de commande automatique de niveau pour un convertisseur analogique-numerique
US4635276A (en) * 1985-07-25 1987-01-06 At&T Bell Laboratories Asynchronous and non-data decision directed equalizer adjustment
EP0238288A1 (fr) * 1986-03-17 1987-09-23 Hewlett-Packard Limited Analyse de transmissions numériques par voie hertzienne

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0465851A2 (fr) * 1990-06-14 1992-01-15 Oy Nokia Ab Récepteur de signaux MAQ qui compense à la fois des distorsions linéaires et non-linéaires
EP0465851A3 (en) * 1990-06-14 1992-08-19 Oy Nokia Ab Qam receiver compensating both linear and non-linear distortion
US5233635A (en) * 1990-06-14 1993-08-03 Oy Nokia Ab Receiving method and receiver for discrete signals
EP0544875A1 (fr) * 1991-06-20 1993-06-09 Universal Data Systems Inc Appareil destine a regler des points de signal, des gains d'egaliseur et autres
EP0544875A4 (en) * 1991-06-20 1993-12-01 Universal Data Systems, Inc. Apparatus for and method of adjusting signal points, equalizer gains and the like
US5428644A (en) * 1992-06-12 1995-06-27 Oy Nokia Ab Adaptive detection method and detector for quantized signals
EP0574223A1 (fr) * 1992-06-12 1993-12-15 Oy Nokia Ab Procédé de détection adaptive et détecteur pour signaux quantifiés
AU662418B2 (en) * 1992-06-12 1995-08-31 Oy Nokia Ab Adaptive detection method and detector for quantized signals
US5680514A (en) * 1994-09-23 1997-10-21 Hughes Electronics Multiple elastic feature net and method for target deghosting and tracking
US5761382A (en) * 1994-09-23 1998-06-02 Hughes Electronics Corporation Multiple elastic feature net and method for target deghosting and tracking
WO1999026423A2 (fr) * 1997-11-13 1999-05-27 Industrial Research Limited Prediction de signaux dans des communications mobiles et systemes utilisant ces predictions pour la reception et le decodage de signaux
WO1999026423A3 (fr) * 1997-11-13 1999-07-15 Ind Res Ltd Prediction de signaux dans des communications mobiles et systemes utilisant ces predictions pour la reception et le decodage de signaux
US10496084B2 (en) 2018-04-06 2019-12-03 Oracle International Corporation Dequantizing low-resolution IoT signals to produce high-accuracy prognostic indicators

Also Published As

Publication number Publication date
FI881007A0 (fi) 1988-03-04
JPH03503234A (ja) 1991-07-18
GB9018361D0 (en) 1990-10-24
AU4074389A (en) 1989-09-22
GB2233865B (en) 1992-09-16
JP2806584B2 (ja) 1998-09-30
GB2233865A (en) 1991-01-16

Similar Documents

Publication Publication Date Title
CA2036992C (fr) Systeme de codage
US5566191A (en) Soft decision maximum likelihood decoding method with adaptive metric
Tsui et al. The construction of smooth models using irregular embeddings determined by a gamma test analysis
US5373456A (en) Expert system for assessing accuracy of models of physical phenomena and for selecting alternate models in the presence of noise
US5315538A (en) Signal processing incorporating signal, tracking, estimation, and removal processes using a maximum a posteriori algorithm, and sequential signal detection
US5216554A (en) Digital phase error estimator
WO1989008360A1 (fr) Procede de detection adaptative pour signaux quantifies
US5588028A (en) Simplified trellis decoder
US4322848A (en) Reliability-weighted analog threshold decoder
US6301296B1 (en) Digital impairment learning sequence
SE521246C2 (sv) Blinddetektion
US4224689A (en) Apparatus for smoothing transmission errors
US5574751A (en) Method for a soft-decision modulation system
US6754293B1 (en) Method and circuit arrangement for processing a signal containing interference
AU662418B2 (en) Adaptive detection method and detector for quantized signals
US6724814B1 (en) Pad and CODEC detection
US5670916A (en) Adaptive equalizer circuit including multiple equalizer units
KR100389775B1 (ko) 단순에러보정을갖는디지탈전송및기록시스템
FI83577C (fi) Adaptivt detekteringsfoerfarande foer kvantiserade signaler.
US5633860A (en) Reliablity-controlled data reception in receivers for TDMA mobile radio systems
US5557638A (en) Transmission system for multivalued digital symbols
US7366470B1 (en) Inter-modulation distortion compensation
Duman et al. Optimal quantization for finite-state channels
US4199722A (en) Tri-state delta modulator
CN111313974A (zh) 一种信号幅度排序与比值计算的调制格式识别方法

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A1

Designated state(s): AU DE GB JP NL US

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): AT BE CH DE FR GB IT LU NL SE

RET De translation (de og part 6b)

Ref document number: 3990156

Country of ref document: DE

Date of ref document: 19910110

WWE Wipo information: entry into national phase

Ref document number: 3990156

Country of ref document: DE

REG Reference to national code

Ref country code: DE

Ref legal event code: 8607