EP0454445A2
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Waveform equalizer using a neural network
Assignee: Fujitsu Ltd
•
Filing: 1991-04-24
•
Granted: —
•
Expires: 2011-04-24
Abstract
Abstract A waveform equalizer for equalizing a distorted signal, contains a sampling unit (10), a time series generating unit (21), and an equalization neural network unit (22). The sampling unit (10) samples the level of a distorted signal at a predetermined rate. The time series generating unit (21) serially receives the sampled level and outputs in parallel a predetermined number of the levels which have been last received. The equalization neural network unit (22) receives the outputs of the time series generating unit, and generates an equalized signal of the distorted signal based on the outputs of the time series generating unit using a set of equalization network weights which are preset therein. The waveform equalizer may further contain a distortion characteristic detecting unit (40), an equalization network weight holding unit (31), and a selector unit (32). The distortion characteristic detecting unit (40) detects a distortion characteristic of the distorted signal. The equalization network weight holding unit (31) holds a plurality of sets of equalization network weights each for being set in the equalization neural network unit (22). The selector unit (31) selects one of the plurality of sets of equalization network weights according to the distortion characteristic which is detected in the distortion characteristic detecting unit (22), and supplies the selected set in the equalization neural network unit (22) to set the selected set therein.
Inventors
- 👤 Takashi Kimoto
- 👤 Kazuo Kawabata
- 👤 Kazuo Asakawa
- 👤 Yasuyuki Ohishi
- 👤 Eisuke Fukuda
- 👤 Takeshi Takano
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-11-06 15:47
CPC Classifications
H; H04; H04L; H04L25/00; H04L25/02; H04L25/03; H04L25/03006; H04L25/03165; G; G06; G06N; G06N3/00; G06N3/02; G06N3/04; G06N3/044; G06N3/0442; G06N3/0499; G06N3/08; G06N3/09; H04L2025/03433; H04L2025/03439; H04L2025/03445; H04L2025/03464
Claims Analysis
Total Claims: 32
Claim 1
Independent Claim
101 words
A waveform equalizer for equalizing a distorted signal, comprising: a sampling means (10) for sampling a level of said distorted signal at a predetermined rate; a time series generating means (21) for serially receiving said sampled level and outputting in parallel a predetermined number of the levels which have been last received; and an equalization neural network means (22) for receiving said outputs of said time series generating means (21), and generating an equalized signal of said distorted signal based on said outputs of said time series generating means (21) using a set of equalization network weights which are preset therein.
Claim 2
Dependent Claim
192 words
A waveform equalizer according to claim 1, wherein said equalization neural network means (22) is a hierarchical network comprising: a predetermined number of input layer units (1-h), a plurality of hidden layer units (1′-i), and an output layer unit (1˝-j); said input layer units (1-h) receiving said outputs of said time series generating means (21), each of said hidden layer units (1′-i) inputting the received values at a predetermined combination of the input layer units (1-h) and outputting a value which is a function of the inputs thereof, and each of said output layer unit (1˝-j)s inputting the outputs of a predetermined combination of the hidden layer units (1′-i) and outputting a value which is a function of the inputs thereof; said equalization network weights are respectively set in the hidden layer units (1′-i) and the output layer units (1˝-j); each of said hidden layer units (1′-i) and said output layer units (1˝-j) comprising a linear combination means (104, 105) for obtaining a linear combination of the inputs thereof using said set of the equalization network weights, and a non-linear function means (106) for obtaining a non-linear function of the linear combination.
Claim 3
Dependent Claim
107 words
A waveform equalizer according to claim 1 or 2, further comprising: a distortion characteristic detecting means (40) for detecting a distortion characteristic of said distorted signal; an equalization network weight holding means (31) for holding a plurality of sets of equalization network weights each for being set in said equalization neural network means (22); and an equalization network weight holding means (32) for selecting one of said plurality of sets of equalization network weights according to the distortion characteristic which is detected in said distortion characteristic detecting means (40), and supplying the selected set to said equalization neural network means (22) to set the selected set therein.
Claim 4
Dependent Claim
107 words
A waveform equalizer according to claim 3, further comprising an optimum equalization network weight obtaining means (50) for obtaining an optimum set of the equalization network weights for each distortion characteristic so that said equalization neural network means (22) outputs a best equalized signal for a distorted signal having the distortion characteristic when the optimum set of the equalization network weights is preset in the equalization neural network means (22), and supplying the obtained set to said equalization network weight holding means (31) so that the equalization network weight holding means (31) stores therein the set of equalization network weights obtained for each of the distortion characteristics.
Claim 5
Dependent Claim
250 words
A waveform equalizer according to claim 4, wherein said optimum equalization network weight obtaining means (50) comprises: a model non-distorted signal storing means (60) for storing a model signal which is not distorted; a model distorted signal generating means for generating a plurality of model distorted signals being respectively generated from said model signal by distorting the model signal according to a plurality of types of distortion characteristics, and supplying the model distorted signals to said equalization neural network means (22); an equalization network weight setting means for setting a set of equalization network weights in said equalization neural network means (22); an output monitoring means for monitoring the output of said equalization neural network means (22); an equalization network weight modifying means for modifying said set of equalization network weights which are set by said equalization network weight setting means in said equalization neural network means (22); a learning control means for controlling said equalization network weight setting means and said equalization network weight modifying means based on successive outputs of said equalization neural network means (22) so that a waveform which is made by the successive outputs of the equalization neural network means (22) accords to the waveform of said model signal, to obtain said optimum set of the equalization network weights for each of said plurality of model distorted signals; and a memorizing means (112 - 121) for memorizing the obtained optimum set of the equalization network weights for each of said plurality of model distorted signals.
Claim 6
Dependent Claim
279 words
A waveform equalizer according to claim 4, wherein said optimum equalization network weight obtaining means (50) comprises: a model non-distorted signal storing means (60) for storing a model signal which is not distorted; a simulated equalization neural network means (22) for receiving model distorted signals, and simulating the operation of said equalization neural network means (22); a model distorted signal generating means for generating a plurality of model distorted signals being respectively generated from said model signal by distorting the model signal according to a plurality of types of distortion characteristics, and supplying the model distorted signals to said simulated equalization neural network means (22); an equalization network weight setting means for setting a set of equalization network weights in said simulated equalization neural network means (22); an output monitoring means for monitoring the output of said simulated equalization neural network means (22); an equalization network weight modifying means for modifying said set of equalization network weights which are set by said equalization network weight setting means in said simulated equalization neural network means (22); a learning control means for controlling said equalization network weight setting means and said equalization network weight modifying means based on successive outputs of said simulated equalization neural network means (22) so that a waveform which is made by the successive outputs of the simulated equalization neural network means (22) accords to the waveform of said model signal, to obtain said optimum set of the equalization network weights for each of said plurality of model distorted signals; and a memorizing means (112 - 121) for memorizing the obtained optimum set of the equalization network weights for each of said plurality of model distorted signals.
Claim 7
Dependent Claim
39 words
A waveform equalizer according to claim 5 or 6, wherein said optimum equalization network weight obtaining means (50) controls the equalization network weight setting means and said equalization network weight modifying means in accordance with a back propagation method.
Claim 8
Dependent Claim
110 words
A waveform equalizer according to any one of claim 3 to 7, wherein said distortion characteristic detecting means (40) comprises: a time series generating means (41) for serially receiving said sampled level and outputting in parallel a predetermined number of the levels which have been last received; a detection neural network means (42) for receiving said outputs of said time series generating means (41) in the distortion characteristic detecting means (40), and generating a distortion characteristic value of said distorted signal based on said outputs of said time series generating means (41) in the distortion characteristic detecting means (40) using a set of detection network weights which are preset therein.
Claim 9
Dependent Claim
26 words
A waveform equalizer according to claim 8, wherein said generation of the distortion characteristic value is carried out during a reception of a predetermined training signal.
Claim 10
Dependent Claim
207 words
A waveform equalizer according to claim 9, wherein said detection neural network means (42) is a hierarchical network comprising: a predetermined number of input layer units (1-h), a plurality of hidden layer units (1′-i), and at least one output layer unit (1˝-j); said input layer units (1-h) receiving said outputs of said time series generating means (41) in the distortion characteristic detecting means (40), each of said hidden layer units (1′-i) inputting the received values at a predetermined combination of the input layer units (1-h) and outputting a value which is a function of the inputs thereof, and each of said output layer unit (1˝-j) inputting the outputs of a predetermined combination of the hidden layer units (1′-i) and outputting a value which is a function of the inputs thereof, and outputting said distortion characteristic value; said detection network weights are respectively set in the hidden layer units (1′-i) and the output layer unit (1˝-j); each of said hidden layer units (1′-i) and each output layer unit (1˝-j) comprising a linear combination means (104, 105) for obtaining a linear combination of the inputs thereof using said set of the detection network weights, and a non-linear function means (106) for obtaining a non-linear function of the linear combination.
Claim 11
Dependent Claim
107 words
A waveform equalizer according to claim 8, 9 or 10, further comprising: an optimum detection network weight obtaining means (50) for obtaining an optimum set of the detection network weights which is common to all distortion characteristics so that said detection neural network means (42) outputs the distortion characteristics corresponding to the distorted signals when said distorted signals after being sampled by the sampling means (10), arc received by the distortion characteristic detecting means (40) when the obtained set is preset in the detection neural network means (42); and a detection network weight presetting means for presetting the obtained set in said detection neural network means (42).
Claim 12
Dependent Claim
179 words
A waveform equalizer according to claim 11, wherein said optimum detection network weight obtaining means (50) comprises: a model distorted signal generating means for generating a plurality of model distorted signals each having a certain distortion characteristic, and supplying the model distorted signals to said detection neural network means (42); a detection network weight setting means for setting a set of detection network weights in said detection neural network means (42); an output monitoring means for monitoring the output of said detection neural network means (42); a detection network weight modifying means for modifying said set of detection network weights which are set by said detection network weight setting means in said detection neural network means (42); a learning control means for controlling said detection network weight setting means and said detection network weight modifying means based on the output of said detection neural network means (42) to obtain said optimum set of the detection network weights; and a presetting means for presetting the obtained optimum set of the detection network weights in said detection neural network means (42).
Claim 13
Dependent Claim
194 words
A waveform equalizer according to claim 11, wherein said optimum detection network weight obtaining means (50) comprises: a model distorted signal generating means for generating a plurality of model distorted signals each having a certain distortion characteristic; a simulated detection neural network means (42) for receiving the model distorted signals, and simulating the operation of said detection neural network means (42); a detection network weight setting means for setting a set of detection network weights in said simulated detection neural network means (42); an output monitoring means for monitoring the output of said simulated detection neural network means (42); a detection network weight modifying means for modifying said set of detection network weights which are set by said detection network weight setting means in said simulated detection neural network means (42); a learning control means for controlling said detection network weight setting means and said detection network weight modifying means based on the output of said simulated detection neural network means (42) to obtain said optimum set of the detection network weights; and a presetting means for presetting the obtained optimum set of the detection network weights in said detection neural network means (42).
Claim 14
Dependent Claim
35 words
A waveform equalizer according to claim 12 or 13, wherein said learning control means controls the detection network weight setting means and said detection network weight modifying means in accordance with a back propagation method.
Claim 15
Dependent Claim
33 words
A waveform equalizer according to any one of claims 3 to 14, in which the distortion characteristic detecting means (40) include means for detecting a plurality of distortion characteristics of the distorted signal.
Claim 16
Dependent Claim
23 words
A waveform equalizer according to claim 15 , wherein said distortion characteristic contains an amplitude distortion, a delay distortion, and a phase distortion.
Claim 17
Dependent Claim
55 words
A waveform equalizer according to claim 15 or 16, when dependent directly or indirectly on claim 8, wherein said distortion characteristic detecting means (40a) includes: a plurality of the said detection neural network means (42a, 42a′, 42a˝), each of the plurality of neural networks generating a value indicating a respective one of said distortion characteristics.
Claim 18
Dependent Claim
159 words
A waveform equalizer according to claim 17, wherein said plurality of detection neural network means (42c-1, 42c-2, 42c-3, 42c-4) each receive said outputs of said time series generating means (41c-1, 41c-2, 41c-3, 41c-4) in the distortion characteristic detecting means (40c), generate and output a value indicating one of said distortion characteristics of said distorted signal, when an amount of the distortion characteristic which is contained in said distorted signal, is within a part, which is assigned to the detection neural network means (42c-1, 42c-2, 42c-3, 42c-4), of the whole range of the distortion characteristic to be detected, and generate and output a predetermined value when said amount of the distortion characteristic is out of said part assigned to the detection neural network means (42c-1, 42c-2, 42c-3, 42c-4), based on said outputs of said time series generating means (41c-1, 41c-2, 41c-3, 41c-4) in the distortion characteristic detecting means (40c) using a set of detection network weights which are preset therein.
Claim 19
Dependent Claim
214 words
A waveform equalizer according to claim 18, further comprising: a plurality of neural network selector means (142), each being provided for one of said distortion characteristics which are detected by said distortion characteristic detecting means (40c), each for receiving the outputs of the plurality of distortion characteristic detecting means (40c) regarding one of the of distortion characteristics, and selecting the output of one of the plurality of distortion characteristic detecting means (40c); and a plurality of selector control means (141), each being provided for one of said plurality of neural network selector means (142), each for receiving the outputs of the plurality of distortion charateristic detecting means (40c) regarding the corresponding one of the distortion one of the plurality of neural network selector means (142) based on the received outputs of the plurality of distortion characteristic detecting means (40c), so that the corresponding neural network selector means (142) selects the output of one of said neural network selector means (142) which does not output said predetermined value, and outputs the maximum value of the output range of said detection neural network means (42c-1, 42c-2, 42c-3, 42c-4) when all of the outputs of the plurality of distortion characteristic detecting means (40c) regarding the corresponding one of the distortion characteristics are equal to the maximum value.
Claim 20
Dependent Claim
72 words
A waveform equalizer according to claim 19, wherein said selector control means (141) sends information on said selection to said equalization network weight holding means (32), and said equalization network weight holding means (32) receives said information together with said output of the neural network selector means (142), and selects the corresponding optimum set of the network weights based on the information and the output of the neural network selector means (142).
Claim 21
Dependent Claim
188 words
A waveform equalizer according to claim 15 or 16, wherein said distortion characteristic detecting means (40b) comprises: a time series generating means (41b, 41b′) for serially receiving said sampled level and outputting in parallel a predetermined number of the levels which have been last received; a plurality of the said detection neural network means (42b, 42b′) each for receiving said outputs of said time series generating means (41b, 41b′) in the distortion characteristic detecting means (40b), generating and outputting values respectively indicating said distortion characteristics of said distorted signal, when each of amounts indicating the distortion characteristics which are contained in said distorted signal, is within a part, which is assigned to the detection neural network means (42b, 42b′), of the whole range of the distortion characteristic to be detected, and generating and outputting a predetermined value when said value indicating the detection characteristic is out of said part assigned to the detection neural network means (42b, 42b′), based on said outputs of said time series generating means (41b, 41b′) in the distortion characteristic detecting means (40b) using a set of detection network weights which are preset therein.
Claim 22
Dependent Claim
215 words
A waveform equalizer according to claim 21, further comprising: a plurality of neural network selector means (142), each being provided for one of said distortion characteristics which are detected by said distortion characteristic detecting means (40b), each for receiving the outputs of the distortion characteristic detecting means (40b) regarding one of the of distortion characteristics, and selecting the output of one of the plurality of detection neural network means (42b, 42b′); and a plurality of selector control means (141), each being provided for one of said plurality of neural network selector means (142), each for receiving the outputs of distortion characteristic detecting means (40b) regarding the corresponding one of the distortion characteristics, and the controlling the corresponding one of the plurality of neural network selector means (142) based on the received outputs of the distortion characteristic detecting means (40b) regarding the corresponding on of the distortion characteristics, so that the corresponding neural network selector means (142) selects the output of one of said neural network means which does not output said predetermined value, and outputs the predetermined value of the output range of said detection neural network means (42b, 42b′) when all of the outputs of the distortion characteristic detecting means regarding the corresponding one of the distortion characteristics are equal to the predetermined value.
Claim 23
Dependent Claim
72 words
A waveform equalizer according to claim 22, wherein said selector control means (141) sends information on said selection to said equalization network weight holding means (32), and said equalization network weight holding means (32) receives said information together with said output of the neural network selector means (142), and selects the corresponding optimum set of the network weights based on the information and the output of the neural network selector means (142).
Claim 24
Dependent Claim
29 words
A waveform equalizer according to claim 22 or 23, wherein said predetermined value is a maximum value of an output range of said detection neural network means (42b, 42b′).
Claim 25
Dependent Claim
29 words
A waveform equalizer according to claim 22, 23 or 24 wherein said generation of the distortion characteristic values is carried out during a reception of a predetermined training signal.
Claim 26
Dependent Claim
118 words
A waveform equalizer according to any one of claims 22 to 25, further comprising: an optimlum detection network weight obtaining means (50) for obtaining an optimum set of the detection network weights which is common to all distortion characteristics so that said plurality of detection neural network means (42b, 42b′) output the distortion characteristics corresponding to the distorted signals when said distorted signals are after being sampled by the sampling means (10), are received by the distortion characteristic detecting means when the obtained set is preset in the plurality of detection neural network means (42b, 42b′); and a detection network weight presetting means for presetting the obtained set in said plurality of detection neural network means (42b, 42b′).
Claim 27
Dependent Claim
112 words
A detector and waveform equalizer for detecting and equalizing a modulated and distorted signal, comprising: a sampling means (10) for sampling a level of said modulated and distorted signal at a predetermined rate; a time series generating means (21) for serially receiving said sampled level and outputting in parallel a predetermined number of the levels which have been last received; and a demodulation/equalization neural network means for receiving said outputs of said time series generating means (21), and generating a demodulated and equalized signal of said modulated and distorted signal based on said outputs of said time series generating means (21) using a set of demodulation/equalization network weights which are preset therein.
Claim 28
Dependent Claim
94 words
A distortion characteristic detector for detecting a distortion characteristic of a distorted signal, comprising: a time series generating means (41) for serially receiving successively sampled levels of said distorted signal and outputting in parallel a predetermined number of the levels which have been last received; a detection neural network means (42) for receiving said outputs of said time series generating means (41), and generating a distortion characteristic value of said distorted signal based on the outputs of said time series generating means (41) using a set of detection network weights which are preset therein.
Claim 29
Dependent Claim
38 words
A detector according to claim 28, including a plurality of the said neural network means (42a, 42a′, 42a"), each of the said means generating a value indicating one of a plurality of distortion characteristics of said distorted signal.
Claim 30
Dependent Claim
119 words
A detector according to claim 28, including a plurality of the said detection neural network means (42), each generating and outputting values respectively indicating said distortion characteristics of said distorted signal, when each of the amounts of the distortion characteristics which are contained in said distorted signal, is within a part, which is assigned to the detection neural network means, of the whole range of the distortion characteristic to be detected, and generating and outputting a predetermined value when said value indicating the detection characteristic is out of said part assigned to the detection neural network means, based on said outputs of said time series generating means (41) using a set of detection network weights which are preset therein.
Claim 31
Dependent Claim
210 words
A distortion characteristic detector according to claim 30, further comprising: a plurality of neural network selector means (142), each being provided for one of said distortion characteristics, each for receiving the outputs of the plurality of detection neural network means regarding one of the of distortion characteristics, and selecting the output of one of the plurality of detection neural network means; and a plurality of selector control means (141), each being provided for one of said plurality of neural network selector means (142), each for receiving the outputs of the plurality of detection neural network means regarding the corresponding one of the distortion characteristics, and the controlling the corresponding one of the plurality of neural network selector means (142) based on the received outputs of the plurality of detection neural network means regarding the corresponding one of the distortion characteristics, so that the selected one of said neural network selector means (142) outputs the value indicating the distortion characteristic which is within said assigned part of the range, and outputs the maximum value of the output range of said detection neural network means when all of the outputs of the plurality of detection neural network means regarding the corresponding one of the distortion characteristics are equal to the maximum value.
Claim 32
Dependent Claim
72 words
A waveform equalizer according to claim 31, wherein said selector control means (141) sends information on said selection to said equalization network weight holding means (32), and said equalization network weight holding means (32) receives said information together with said output of the neural network selector means (142), and selects the corresponding optimum set of the network weights based on the information and the output of the neural network selector means (142).
References Cited (Backward Citations)
No citation records indexed.
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.