JPH09160895A - Wavelength multiplexed light receiver - Google Patents

Abstract

PROBLEM TO BE SOLVED: To correctly execute the learning of a light receiver even when the distance between the light receiver and a light transmitter is large by adopting a reception reproducing teacher signal as a teacher signal so as to make the relative phase of the teacher signal normally coincide with the relative phase of a reception learning signal. SOLUTION: Signals having mutually different speeds by the repetition pattern of one and zero are generated as the wavelength-corresponding teacher signal from a teacher signal generator 31 on the transmission side so as to transmit wavelength-multiplexed light. On the reception side, wavelength-multiplexed light from an optical fiber 1 is branched and transduced into an electric signal by a photodetector 45. Then, the basic frequency components of the respective teaching signals are separated by filters 461 -464 and supplied to a learning circuit 23 as the teacher signal of the one and zero pattern. The other branced light is wavelength-separated, transduced into the electric signal by photodetector array 7 and supplied to a neural network 9.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optical receiver having a learning means for separating a modulated signal for each wavelength from a wavelength multiplexed light by using a neural network, and for allowing the neural network to learn its weight and threshold value. It is related to vessels.

[0002]

2. Description of the Related Art In optical communication, there is a so-called wavelength multiplexing technique for transmitting optical signals of a plurality of wavelengths at the same time and assigning different information to each wavelength in order to effectively utilize a transmission line. This technique requires an optical receiver having a function of separating an optical signal for each wavelength. First, the operation of this wavelength division multiplexing optical receiver will be described with reference to FIG. This optical receiver is shown in Japanese Patent Application No. 2-101627. In FIG. 3, wavelength-multiplexed signal light 3 of wavelengths λ ₁ , λ ₂ , ..., λ _n propagated from the optical fiber 1 through the incident side lens 2 is made incident on a bulk type diffraction grating 4 and is diffracted here. The light 5 in which the respective wavelengths are angularly dispersed is incident on the light receiving element array 7 via the exit side lens 6. That is, the wavelength multiplexed signal light 3
Is wavelength-separated by the wavelength separation means, in this example, the diffraction grating 4, and the spatial distribution of the wavelength-separated light is incident on the light-receiving element array 7 as the photoelectric conversion means.

In the light receiving element array 7, the light is diffracted.
Light 5 has each wavelength λ₁, Λ_Two，… ， Λ _nDifferent light receiving element for each
Two-dimensional light with peak light intensity at the child array position
Converted to intensity distribution. Obtained from the light receiving element array 7
An electrical signal 8 having a current distribution corresponding to the two-dimensional light intensity distribution
To the neural network 9 and output its output port.
Output signal (modulation signal) corresponding to each wavelength from G.10
O₁, O_Two, ..., O_nGet. For each wavelength
Electrical signals can be obtained from different elements in Ray 7, but separated
The tails of adjacent light distributions of each
It was not always separated accurately from the element array 7 for each wavelength.
It is difficult to obtain an electrical signal, and the source wave on the transmitting side
Each element of the light-receiving element array 7 and wavelength due to fluctuations in length, etc.
Relationship with is not constant. From this point, neural network
Network 9 is used.

As a learning method of a neural network in this optical receiver, a proposal shown in Japanese Patent Application No. 6-68582 has been made. The learning device will be described with reference to FIG. Here, parts corresponding to those in FIG. 3 are designated by the same reference numerals. The teacher signal generator 11 provided on the transmission side can generate the synchronization signal Ts and the teacher signal, and first outputs the synchronization signal Ts. Synchronizing signal Ts is inputted to the laser driver 12 _0. Laser driver 12 _{0-12
At 4} , the drive signal for driving the laser diode 13 (LD0 to LD4) is generated by the output signal of the teacher signal generator 11. The synchronization signal Ts is transmitted as a signal (wavelength λ ₀ ) of the laser diode LD0. The light of wavelength λ ₀ modulated by the synchronization signal Ts is isolated by the isolator 14 and the optical demultiplexer 15.
After being sequentially passed through, the light is incident on the optical fiber 1. Light emitted from the optical fiber 1 is converted into a light converting means 18 (lens 2 in FIG. 3,
After being wavelength-separated by the diffraction grating 4 or the like), the light passes through the lens 6 and is received by the light receiving element array 7 as described in FIG.

An optical signal of wavelength λ ₀ is previously received by the light receiving element array 7
It is assumed that the element PD0 is set to receive light. Element PD of light receiving element array 7
The output electric signal from 0 is amplified by the preamplifier array 21 and guided to the teacher signal generator 24. Teacher signal generator 24
Then, the synchronization signal Ts sent from the teacher signal generator 11
Teacher signal generation unit 11 that extracts the
Generates the same teacher signal as the teacher signal generated in. That is, the teacher signal generator 24 on the receiving side has a built-in circuit that generates the same teacher signal as the teacher signal generator 11 on the transmitter side, and synchronizes with the synchronization signal Ts from the teacher signal generator 11. And generate a teacher signal. Teacher signal generator 2
The teacher signal output from 4 is input to the learning circuit 23.

On the other hand, the teacher signal generator 11 on the transmitting side is synchronized.
After transmitting the signal Ts, the teacher signal is transmitted. Teacher signal
Is the laser driver 12₁~ 12_FourIs input to laser
Driver 12₁~ 12_FourThen, the output of the teacher signal generator 11
Laser diode 13 by force signal₁~ 13_Four(LD1
~ LD4) drive signal for driving. Four rays
The Diode 13₁~ 13_FourFrom (LD1-LD4)
The output light passes through the isolator 14 and is then combined with the light.
Combined by wave filter 15 and incident on one optical fiber 1.
You. Here, the laser diode 13₁~ 13_FourEach of
The wavelength of the emitted light is λ ₁, Λ_Two, Λ_Three, Λ_FourAnd Light fa
The light emitted from the aver 1 passes through the light conversion means 18 and the lens 6.
After that, the light is received by the light receiving element array 7. Photodetector array
The output signal from 7 is amplified by the preamplifier array 21.
Is input to the neural network 9. Nura
In the network 9, internal parameters (synaptic load
Weight, threshold value) and output port 10
Output signal. Also the neural network 9
The output signal of is also branched and supplied to the learning circuit 23. Learning times
In path 23, the output signal from the neural network 9
And learning using the teacher signal from the teacher signal generator 24
I do. In the learning circuit 23, the teacher signal and the neural network
To reduce the error sum of squares with the output signal of work 9
Is an internal parameter of the neural network 9.
The synaptic weight and threshold are changed sequentially, and the error sum of squares
Learning is considered completed when
The learning process ends. As the learning circuit 23 described above, the present application
Learning based on the multi-frequency vibration learning method proposed by humans
Circuits can be used (references, 1992 Electronics
IEICE Spring National Convention, SD-2-6, Multi-frequency vibration
Analog Neural Network High Speed Based on Learning Method
Learning experiment).

FIG. 5 shows the teacher signal generator 1 on the transmitting side in FIG.
1 synchronization signal and learning signal (teacher signal)
Here is an example. First, the teacher signal generator 11 first outputs the synchronization signal Ts.
Is generated for a certain period of time, and then the teacher signal T₁~ T_FourSend out
I do. Teacher on the receiving side while the synchronization signal Ts is being transmitted
The signal generation unit 24 synchronizes the signals and ₁
~ T_FourIs transmitted, the sender's teaching
The same teacher signal as the teacher signal sent from the teacher signal generator 11.
Signal is generated and the teacher signal is input to the learning circuit 23.

FIG. 6 shows a conventional optical receiver after learning is completed.
Shows the operation of. 6 is different from FIG. 4 in that
The point that the laser diode LD0 is not required
Output to the teacher signal generator 24 and its learning circuit 23
Force, to the learning circuit 23 of the neural network 9
The point is that there is no output. After learning, with each wavelength of light
The data signals Data1 to Data4 you want to send are
Driver 12₁~ 12_FourTo enter. Laser driver 12
₁~ 12_FourOutput of laser diode 13 ₁~ 13
_FourAre driven respectively, and each laser diode (LD1 to LD1
Light modulated by Data1 to Data4 from LD4)
A signal is output, and these optical signals are combined by the multiplexer 15.
And sent to the optical fiber 1. On the other hand, in the optical receiver 17,
The wavelength-multiplexed light output from the optical fiber 1 is the optical conversion means 1.
8. After passing through the lens 6, the light receiving element array 7 converts it into an electrical signal.
To be converted. The output signal from the light receiving element array 7 is
After being amplified by the amplifier array 21, the neural network is
It is input to network 9. In the neural network 9,
The appropriate synaptic weights and thresholds obtained by learning are
It is added from the learning circuit 23. The input signal is
It is processed according to the naps load and the threshold,
As a result, Data1 to Data4 are neural networks.
It is output from each of the output ports 10 of 9.

[0009]

Since optical fibers have different dispersion values depending on wavelengths, when the optical fiber length reaches several tens of km, the received teacher signals are out of phase with each other. . However, in the conventional learning method, the same signal as the teacher signal generated in the teacher signal generating section on the transmitting side is generated in the teacher signal generating section on the receiving side based on the synchronization signal Ts. There is a possibility that the relative phase and the relative phase between the received teacher signals do not match and learning of the optical receiver is not performed correctly.

[0010]

In the present invention, the teacher signal generating means on the transmitting side uses fixed patterns of "0101 ..." As the teacher signals, the speeds of which are different from each other, and light of each wavelength is modulated by each teacher signal. On the side of the wavelength-multiplexed optical receiver of the present invention, the fundamental frequency component of each teacher signal is extracted from the received wavelength-multiplexed light as an electric signal by the bandpass filter, and these signals are extracted by the waveform shaping means as necessary. The waveform is shaped into 0101 ... And the teacher signal is reproduced, and this reproduced teacher signal is supplied to the learning means as a teacher signal.

To reproduce the teacher signal, the received wavelength-multiplexed light is branched, and the branched light is collectively converted into an electric signal by the photoelectric conversion means, and each fundamental frequency component is extracted from the electric signal by the filter means. . Or, in order to extract the respective modulated signals from the received wavelength-multiplexed light, the wavelengths are separated and the wavelength-separated light is converted into an electric signal by the light receiving element array.
The signals supplied to the neural network of each element (element) of this light receiving element array are branched and added by the adding means, and the added signal is separated into each fundamental frequency component by the filter means.

With such a configuration, the reproduction teacher signal holds the relative phase difference due to the wavelength received by the dispersion of the transmission optical fiber, which is the dispersion value of the transmission optical fiber at the time of receiving the communication information (data). It becomes the same as the relative phase difference based on the difference due to the wavelength, that is, the relative phase of the teacher signal and the relative phase of the learning signal always match, the neural network is correctly learned, and the modulated signal can be demodulated accurately.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows a first embodiment of the present invention, in which parts corresponding to those in FIG. The teacher signal generation unit 31 on the transmission side outputs each output (T1, T2,
A fixed pattern of "0101 ..." With different signal speeds is output to T3 and T4). Laser driver 12 _{1 to} 12 ₄
Then, a drive signal for driving the laser diodes 13 _{1 to} 13 ₄ (LD1 to LD4) is generated by each output signal of the teacher signal generation unit 31. Four laser diodes 13 ₁ ~
Output light from 13 ₄ (LD1 to LD4) is
After passing through the isolator 14, it is multiplexed by the optical multiplexer 15.
It is incident on one optical fiber 1. Here, the wavelengths of the emitted lights of the laser diodes 13 ₁ , 13 ₂ , 13 ₃ , and 13 ₄ are λ ₁ , λ ₂ , λ ₃ , and λ ₄ , respectively. The wavelength division multiplexed light received from the optical fiber 1 is branched into two by the optical branching circuit 48, and one outgoing light is incident on the teacher signal generator 44. The light incident on the teacher signal generator 44 receives the light receiving element 45.
Is converted into an electric signal by the filter, and the electric signal is sent to the band pass filters 46 _{1 to} 46 ₄ (BPF _{1 to} BPF ₄ ) and the fundamental frequency components of the teacher signals are extracted by the filters 46 _{1 to} 46 ₄ . Each of these extracted signal waveform shaper 49 _{1-49 4} due respectively thresholding "1", "0" repeating pattern to become the waveform-shaped the teacher signal reproduced by the learning circuit 23 Supplied. Here, the output signal of the bandpass filter 46 may be supplied to the learning circuit 23 as a teacher signal without waveform shaping. However, if the waveform is shaped, the reproducing ability can be improved.

The light emitted from the other side of the light branching circuit 48 enters the light receiver 17, passes through the light converting means 18 and the lens 6, and is received by the light receiving element array 7. Light receiving element array 7
Then, the optical signal is converted into an electric signal, and the output signal thereof is amplified by the preamplifier array 21 as necessary and input to the neural network 9. Neural network 9
Then, processing is performed according to internal parameters (synapse weight, threshold value), and an output signal is output from the output port 10. The output signal of the neural network 9 is the learning circuit 2
3 is branched and input. In the learning circuit 23, the output signal from the neural network 9 and the teacher signal generator 44
Learning is performed using the reproduced teacher signal from. Learning circuit 23
Then, the synapse weight and the threshold value, which are internal parameters of the neural network 9, are sequentially changed so that the sum of squared errors between the teacher signal and the output signal of the neural network 9 becomes smaller, and the sum of squared errors falls below a certain set value. If so, it is considered that the learning has ended, and the learning process ends. As the learning circuit 23 described above, a learning circuit based on the multi-frequency vibration learning method proposed by the present applicant can be used (reference document, 1992 IEICE Spring National Convention, SD-2-6, Multi. Analog neural network high-speed learning experiment based on frequency vibration learning method).

The speed difference between the teacher signals corresponding to the respective wavelengths transmitted on the transmitting side may be such that they can be easily separated by the filters 46 _{1 to} 46 ₄ on the receiving side, and the speed at which normal information (data) is transmitted is 150 Mb / In the case of s, the speed of each teacher signal is 2 Mb / s, 1 Mb / s, 0.5 Mb / s, 0.25 M
b / s or the like. FIG. 2 shows a second embodiment of the present invention, and the same reference numerals are given to the portions corresponding to those in FIG. On the transmitting side of the second embodiment, the same optical signal as in the first embodiment is incident on the optical fiber 1. The light emitted from the optical fiber 1 enters the optical receiver 17, passes through the optical conversion means 18 and the lens 6, is converted into an electric signal by the light receiving element array 7, and the output signal from the light receiving element array 7 is
If necessary, the signal is amplified by a compact preamplifier array and input to the neural network 9 and at the same time, to the teacher signal generation unit 44. The respective signals input to the teacher signal generator 44 are added by the adder 47, and then the fundamental frequency components of the respective teacher signals are extracted by the bandpass filters 46 _{1 to} 46 ₄ (BPF1 to BPF4). The extracted signal is supplied to the learning circuit 23 as a teacher signal.
This output signal of the band pass filter 46 ₁ to 46 ₄ may the waveform shaper 49 _{1-49 4,} it is possible to improve the reproduction capability by supplying to the waveform shaping to the learning circuit 23. Others are the same as the first embodiment.

[0016]

As described above, according to the present invention, the received reproduction teacher signal is used as the teacher signal. Therefore, since the relative phase of this teacher signal and the relative phase of the received learning signal always match, the optical receiver and the optical receiver Even if the distance between the transmitters is long, the learning of the optical receiver can be correctly performed and the data can be correctly demodulated.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a block diagram showing another embodiment of the present invention.

FIG. 3 is a block diagram showing a conventional optical receiver.

FIG. 4 is a block diagram showing a conventional optical receiver with a learning function.

FIG. 5 is a waveform diagram showing a conventional generated teacher signal on the transmission side.

FIG. 6 is a block diagram for explaining the operation of a conventional optical receiver after learning.

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] December 7, 1995

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Claim 1

[Correction method] Change

[Correction contents]

[Procedure amendment 2]

[Document name to be amended] Statement

[Correction target item name] Claim 2

[Correction method] Change

[Correction contents]

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI technical display location H04B 10/14 10/04 10/06

Claims (4)

[Claims] 1. The received wavelength-multiplexed light is spatially separated by the wavelength separation means for the difference in the wavelength of the light, and the spatial distribution of the wavelength-separated light is converted into an electric signal by the photoelectric conversion means, which is then converted. The signal is processed by the electric signal neural network and the modulated signal of each wavelength light is separated and taken out. The teacher signal and the output of the neural network are input to the learning means, and the weight and the threshold are set in the neural network. In the wavelength-division-multiplexing optical receiver to be learned, the wavelength-division-multiplexing light is modulated by a teacher signal having a repeating fixed pattern of "1" and "0" whose speeds are different from each other. Each teacher signal is reproduced by the teacher signal reproducing means, and the learning means learns by learning the reproduced teacher signal and the output of the neural network. A wavelength division multiplexing optical receiver characterized by performing learning operation. 2. The teacher signal reproducing means, means for branching the received wavelength-multiplexed light, means for simultaneously converting each wavelength of the branched wavelength-multiplexed light into an electric signal, and the converted electric signal Filter means for extracting the fundamental frequency components of each of the teacher signals in parallel, shaping the extracted fundamental frequency components into a repeating pattern of "1" and "0" of the frequencies to obtain the reproduced teacher signal, and The wavelength division multiplexing optical receiver according to claim 1, characterized in that 3. The teacher signal reproducing means parallelly adds the fundamental frequency component of each teacher signal from the output of the adding means and the adder means to which the converted electrical signals from the photoelectric conversion means are branched and added. 2. The wavelength division multiplexing optical receiver according to claim 1, further comprising a filter means for taking out the reproduced teacher signal. 4. A waveform shaping means for extracting the fundamental frequency component from the filter means and shaping the fundamental frequency component into a repeating pattern of "1" and "0" of the frequency to obtain the reproduction teacher signal. The wavelength division multiplexing optical receiver according to claim 2 or 3.