CN206480025U - Pattern classifier based on RBF - Google Patents
Pattern classifier based on RBF Download PDFInfo
- Publication number
- CN206480025U CN206480025U CN201620745796.8U CN201620745796U CN206480025U CN 206480025 U CN206480025 U CN 206480025U CN 201620745796 U CN201620745796 U CN 201620745796U CN 206480025 U CN206480025 U CN 206480025U
- Authority
- CN
- China
- Prior art keywords
- transistor
- colelctor electrode
- base stage
- stage
- output end
- Prior art date
- Legal status (The legal status 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 status listed.)
- Expired - Fee Related
Links
Landscapes
- Amplifiers (AREA)
Abstract
The utility model provides a kind of pattern classifier based on RBF, and it includes first to fourth RBF neuron circuits module, first to fourth resistance and the first to the 5th Gilbert multipliers;The current output terminal of described first to the 5th Gilbert multipliers is sequentially connected the output end I as graderout;The output of first RBF neuron circuits connects the first input of first resistor one end and the first Gilbert multipliers respectively;The first resistor other end is grounded respectively and the first Gilbert multipliers second are inputted;The output of 2nd RBF neuron circuits connects the first input of second resistance one end and the 2nd Gilbert multipliers respectively;The second resistance other end is grounded respectively and the 2nd Gilbert multipliers second are inputted;The output of 3rd RBF neuron circuits connects 3rd resistor one end respectively and the 4th Gilbert multipliers first are inputted;The 3rd resistor other end is grounded respectively and the 4th Gilbert multipliers second are inputted;The output of 4th RBF neuron circuits connects the 4th resistance one end and the 5th Gilbert multipliers respectively;The 4th resistance other end is grounded respectively and the 5th Gilbert multipliers second are inputted.The grader with having the advantages that Residuated Lattice, it is portability, high speed, expansible.
Description
Technical field
The utility model is related to a kind of pattern classifier, and in particular to a kind of pattern classifier based on RBF.
Background technology
The theoretical model of RBF (RBF, Radial Basic Function) neutral net is in pattern classification, letter
Number approach wait artificial intelligence field be widely used, but at present be also concentrated mainly on traditional computer software simulate
In realization.Realization of the RBF neutral nets on software is all to use universal cpu processor, it has not been convenient to be embedded into other application system
In system, and study computing is completed by the general-purpose computing system of bulky, do not possess portability.In calculating process
In, CPU be often RBF to be waited until neuron calculated one by one after, then calculate the long and, use
Serial computing mode, speed is slower.Therefore, the software of RBF neural, which is realized, is difficult to meet it in artificial intelligence application field
At a high speed, it is portable, can be embedded in terms of requirement.
The hardware of RBF neural is realized, special neural network chip can be integrated into, with small volume, carrying side
Just the characteristics of, easily it is embedded into other systems and realizes special function.In addition, it can also realize the parallel computation of height, gram
The defect that RBF neural is realized on software is taken.Therefore, the Research of Hardware Implementation of RBF neural is significant.
The content of the invention
It is suitable by parallel connection the utility model proposes a kind of analog circuit implementation of the pattern classifier based on RBF
When the pattern classifier of quantity, appropriate extraneous bias voltage is given, can be achieved to reflect the point that can not be divided in a space
Another space is mapped to, so that the point that script can not be divided carries out simply linear partition, the function of implementation pattern classification.
The utility model is achieved through the following technical solutions:A kind of pattern classifier based on RBF, it is characterised in that:Bag
Include first to fourth RBF neuron circuits module, first to fourth resistance and the first to the 5th Gilbert multipliers;Described
The current output terminal of one to the 5th Gilbert multipliers is sequentially connected the output end I as graderout;First RBF neurons
The output of circuit connects the first input of first resistor one end and the first Gilbert multipliers respectively;The first resistor other end is distinguished
Ground connection and the first Gilbert multipliers second are inputted;The output of 2nd RBF neuron circuits connect respectively second resistance one end and
First input of the 2nd Gilbert multipliers;The second resistance other end is grounded respectively and the 2nd Gilbert multipliers second are defeated
Enter;The output of 3rd RBF neuron circuits connects 3rd resistor one end respectively and the 4th Gilbert multipliers first are inputted;3rd
The resistance other end is grounded respectively and the 4th Gilbert multipliers second are inputted;The output of 4th RBF neuron circuits connects respectively
Four resistance one end and the 5th Gilbert multipliers;The 4th resistance other end is grounded respectively and the 5th Gilbert multipliers second are defeated
Enter;First output end of the first RBF neuron circuit modules is Vx, the second output end is Vx0, the 3rd output end is Vy, the
Four output ends are Vy0;First output end of the 2nd RBF neuron circuit modules is Vx, the second output end is Vx0', the 3rd
Output end is Vy, the 4th output end is Vy0';First output end of the 3rd RBF neuron circuit modules is Vx, the second output
Hold as Vx0", the 3rd output end is Vy, the 4th output end is Vy0”;First output end of the 4th RBF neuron circuit modules
For Vx, the second output end is Vx0" ', the 3rd output end is Vy, the 4th output end is Vy0”';First Gilbert multipliers first are defeated
It is V to go out endw1, the second output end is Vw2, the 2nd the first output end of Gilbert multipliers is Vw1', the second output end is Vw2';The
Three the first output ends of Gilbert multipliers are Vw1", the second output end is Vw2”;4th the first output end of Gilbert multipliers is
Vw1" ', the second output end is Vw2”';5th the first output end of Gilbert multipliers is Vw1" ", the second output end Vw2" ", the 3rd
3rd end of Gilbert multipliers is Vb。
In the embodiment of the utility model one, Gilbert multipliers include the first to the 17th transistor M1~M17;The
One transistor to the 6th transistor M1~M6Emitter stage link together and connect high level;The first transistor M1Base stage connect second
Transistor M2Base stage;The first transistor M1Colelctor electrode connect and meet the 9th transistor M respectively9Emitter stage and the tenth transistor M10
Emitter stage;Second transistor M2Base stage connect the colelctor electrode of second transistor;Second transistor M2Colelctor electrode connect the 7th crystal
Pipe M7Colelctor electrode;Third transistor M3Base stage meet the 4th transistor M4Base stage;Third transistor M3Base stage connect the 3rd
Transistor M3Colelctor electrode;Third transistor M3Colelctor electrode meet the 8th transistor M8Colelctor electrode;4th transistor M4Collection
Electrode pole meets the 11st transistor M respectively11Emitter stage and the tenth two-transistor M12Emitter stage;5th transistor M5Base stage
Meet the 6th transistor M6Base stage;5th transistor M5Base stage meet the 5th transistor M5Colelctor electrode;5th transistor M5Collection
Electrode meets the 14th transistor M14Colelctor electrode;6th transistor M6Colelctor electrode meet the 17th transistor M respectively17Colelctor electrode
And output Iout1;7th transistor M7Base stage meet Vw1;7th transistor M7Emitter stage connect the emitter stage of the 8th transistor respectively
And the 13rd transistor M13Colelctor electrode;8th transistor M8Base stage meet Vw2;9th transistor M9Base stage to connect the 12nd brilliant
Body pipe M12Base stage;9th transistor M9Colelctor electrode meet the 11st transistor M respectively11Colelctor electrode and the 15th transistor
M15Colelctor electrode;Tenth transistor M10Base stage meet the 11st transistor M11Base stage;Tenth transistor M10Colelctor electrode difference
Meet the tenth two-transistor M12Colelctor electrode and the 16th transistor M16Colelctor electrode;9th transistor M9With the tenth transistor M10
Base stage meet V respectivelyinBoth positive and negative polarity;13rd transistor M13Base stage meet bias voltage Vbias;14th transistor M14Base
The transistor M of pole the 15th15Base stage;15th transistor M15Base stage connect its colelctor electrode;16th transistor M16Base stage connect
17th pipe M17Base stage;16th transistor M16Base stage connect its colelctor electrode;The emitter stage of 13rd to the 17th transistor connects
Connect and be grounded together.
In the embodiment of the utility model one, the sqrt circuit includes the 18th to the 26th transistor M18~
M26;The 18th transistor M18Colelctor electrode and the 19th transistor M19Base stage connection is used as sqrt circuit input end Iin;
18th transistor M18Base stage meets the 26th transistor M respectively26Base stage, the 19th transistor M19Emitter stage and the 24th
Transistor M24Colelctor electrode;19th transistor M19Base stage meets the 21st transistor M21Base stage;20th transistor M20Base stage
Connect its colelctor electrode;21st transistor M21Emitter stage meets the 20th transistor M20Colelctor electrode;21st transistor M21Current collection
Pole meets the 20th two-transistor M22Colelctor electrode;20th two-transistor M22Base stage connects the 23rd transistor base;19th is brilliant
Body pipe M19Colelctor electrode, the 20th two-transistor M22Emitter stage and the 23rd transistor M23Emitter stage, which links together, connects high electricity
It is flat;23rd transistor M23Colelctor electrode meets the 25th transistor M respectively25Colelctor electrode and output Iout2;24th crystal
Pipe M24Base stage meets the 25th transistor M respectively25Base stage and output end Vb;26th transistor M26Colelctor electrode connects output
Iout2;18th transistor M18Colelctor electrode and the 24th the 26th transistor M24~M26Emitter stage links together ground connection.
In the embodiment of the utility model one, it is brilliant that the class Gaussian function generation circuit includes the 27th to the 52nd
Body pipe M27~M52;27th transistor M27Emitter stage, the 28th transistor M28Emitter stage and the 35th to the 4th
Ten two-transistor M35~M42Emitter stage, which links together, connects high level;The 27th transistor M27Base stage connects second respectively
17 transistor M27Colelctor electrode and the 28th transistor M28Base stage;27th transistor M27Colelctor electrode connect the 20th
Nine transistor M29Emitter stage;28th transistor M28Colelctor electrode meets the 30th transistor M30Emitter stage;29th is brilliant
Body pipe M29Base stage meets the 29th transistor M respectively29Colelctor electrode meets the 30th transistor M30Base stage;29th transistor M29
Colelctor electrode meets the 31st transistor M31Colelctor electrode;30th transistor M30Colelctor electrode meets the 47th transistor M47Colelctor electrode;
31st transistor M31Colelctor electrode connects its base stage;31st transistor M31Base stage meets the 31st transistor M respectively31Collection
Electrode and the 30th two-transistor M32Base stage;31st transistor M31Emitter stage meets the 33rd M33Colelctor electrode;32nd
Transistor M32Colelctor electrode meets the 35th M35Colelctor electrode;30th two-transistor M32Emitter stage meets the 34th transistor M34Collection
Electrode;33rd transistor M33Base stage meets the 33rd transistor M respectively33Colelctor electrode and the 34th transistor M34Base stage;
33rd transistor M33Emitter stage, the 34th transistor M34Emitter stage, the 50th to the 50th two-transistor emitter stage connect
Be connected together ground connection;35th transistor M35Base stage connects the 35th transistor M respectively35Colelctor electrode, the 36th crystal
Pipe M36Base stage, the 41st transistor M41Base stage and the 40th two-transistor M42Base stage;36th transistor M36Colelctor electrode connects
37th transistor M37Base stage;37th transistor M37Colelctor electrode meets the 43rd transistor M respectively43Colelctor electrode and connect
45th transistor M45Colelctor electrode;37th transistor M37Base stage meets the 38th transistor M38Base stage;38th is brilliant
Body pipe M38Base stage, the 39th transistor M39Colelctor electrode, the 40th two-transistor M42Colelctor electrode, which links together, connects output
Iout3;39th transistor M39Base stage meets the 40th transistor M respectively40Base stage and the 41st transistor M41Colelctor electrode;The
40 transistor M40Colelctor electrode meets the 44th transistor M respectively44Colelctor electrode and the 46th transistor M46Colelctor electrode;4th
13 transistor M43Base stage and the 46th transistor M46Base stage meets input V togetherin, the 43rd transistor M43Emitter stage point
The 44th transistor M is not met44Emitter stage and the 48th transistor M48Colelctor electrode;44th transistor M44Base stage meets V1;
45th transistor M45Base stage meets V2;45th transistor M45Emitter stage meets the 46th transistor M respectively46Emitter stage
And the 49th transistor M49Colelctor electrode;47th transistor M47Base stage distinguishes the 47th transistor M47Colelctor electrode, the 4th
18 transistor M48Base stage, the 49th transistor M49Base stage;47th transistor M47Emitter stage meets the 50th transistor M50
Colelctor electrode;48th transistor M48Colelctor electrode meets the 51st transistor M51Colelctor electrode;49th transistor M49Transmitting
Pole meets the 50th two-transistor M52Colelctor electrode;50th transistor M50Base stage meets the 50th transistor M respectively50Colelctor electrode, the 5th
11 transistor M51Base stage, the 50th two-transistor M52Base stage.
It is suitable by parallel connection the utility model proposes a kind of analog circuit implementation of the pattern classifier based on RBF
When the pattern classifier of quantity, appropriate extraneous bias voltage is given, the function of pattern classification can be achieved.The utility model can collect
As special neural network chip, there is small volume, be convenient for carrying, can be embedded in, it is possible to achieve the parallel meter of height
Calculate, the volume for overcoming software implementation pattern grader is big, not portable, be difficult the defect that is embedded in, arithmetic speed is slow.This practicality
It is new can also be by suitably increasing the number of RBF neuron circuit modules or by the way that the utility model to be carried out to side in parallel
Formula, carrys out the function of expanded circuit, it is solved more complicated pattern classification problem.The utility model is by its Residuated Lattice, just
The property taken, high speed, it is expansible the advantages of, be expected to be widely used in artificial intelligence fields such as pattern classifications.
Brief description of the drawings
Fig. 1 is the schematic diagram of pattern classifier.
Fig. 2 is the schematic diagram of pattern classifier.
Fig. 3 is the transistor-level schematic of Gilbert multipliers.
Fig. 4 is the transistor-level schematic of sqrt circuit.
The transistor-level schematic of Fig. 5 class Gaussian function generation circuits.
The analogous diagram of Fig. 6 class Gaussian function generation circuits.
The simulation waveform of Fig. 7 class Gaussian function generation circuits and ideal Gaussian function comparison diagram.
The distribution map of five points of Fig. 8.
The functional simulation figure of Fig. 9 pattern classifiers.
Embodiment
The utility model is described further with reference to the accompanying drawings and detailed description.
The utility model utilizes Gilbert multipliers, sqrt circuit, and these are substantially electric for class Gaussian function generation circuit
Road unit devises a pattern classifier based on RBF.As shown in figure 1, the pattern classifier has two input (Vx, Vy),
One output end (Iout), and 21 control end (wherein (Vx0, Vy0), (Vx0', Vy0'), (Vx0", Vy0"), (Vx0" ',
Vy0" ') respectively represent four class Gaussian functions center (i.e. control input-hidden layer weights), V1And V2For controlling class high
The shape (i.e. control input-hidden layer threshold value) of this function, (Vw1, Vw2), (Vw1', Vw2'), (Vw1", Vw2"), (Vw1" ',
Vw2" '), (Vw1" ", Vw2" ") it is respectively intended to (V in five implicit-output layer weights of control, Fig. 2w1", Vw2") control it is implicit-
1) output layer weights are defaulted as, VbFor controlling to imply-output layer threshold value).By loading appropriate bias voltage in control end,
Just the function that several points on two dimensional surface are correctly classified can be realized.The data of point i.e. on two dimensional surface are by (Vx, Vy)
Some points in plane are classified as Type1 by input, the pattern classifier, now network output IoutFor 1;And by plane other one
A little points are classified as Type2, now network output IoutFor 0.
Schematic diagram of the present utility model as shown in Fig. 2 the grader include first to fourth RBF neuron circuits module,
First to fourth resistance and the first to the 5th Gilbert multipliers;The electric current output of described first to the 5th Gilbert multipliers
End is sequentially connected the output end I as graderout;The output of first RBF neuron circuits connects first resistor one end and respectively
First input of one Gilbert multipliers;The first resistor other end is grounded respectively and the first Gilbert multipliers second are inputted;
The output of 2nd RBF neuron circuits connects the first input of second resistance one end and the 2nd Gilbert multipliers respectively;Second electricity
The resistance other end is grounded respectively and the 2nd Gilbert multipliers second are inputted;The output of 3rd RBF neuron circuits connects the 3rd respectively
Resistance one end and the 4th Gilbert multipliers first are inputted;The 3rd resistor other end is grounded and the 4th Gilbert multiplication respectively
Device second is inputted;The output of 4th RBF neuron circuits connects the 4th resistance one end and the 5th Gilbert multipliers respectively;4th
The resistance other end is grounded respectively and the 5th Gilbert multipliers second are inputted;The first of the first RBF neuron circuit modules
Output end is Vx, the second output end is Vx0, the 3rd output end is Vy, the 4th output end is Vy0;The 2nd RBF neurons electricity
First output end of road module is Vx, the second output end is Vx0', the 3rd output end is Vy, the 4th output end is Vy0';Described
First output end of three RBF neuron circuit modules is Vx, the second output end is Vx0", the 3rd output end is Vy, the 4th output end
For Vy0”;First output end of the 4th RBF neuron circuit modules is Vx, the second output end is Vx0" ', the 3rd output end is
Vy, the 4th output end is Vy0”';First the first output end of Gilbert multipliers is Vw1, the second output end is Vw2, second
The output end of Gilbert multipliers first is Vw1', the second output end is Vw2';3rd the first output end of Gilbert multipliers is
Vw1", the second output end is Vw2”;4th the first output end of Gilbert multipliers is Vw1" ', the second output end is Vw2”';5th
The output end of Gilbert multipliers first is Vw1" ", the second output end Vw2" ", the 3rd end of the 3rd Gilbert multipliers is Vb。
Fig. 3 is the transistor-level schematic of Gilbert multipliers, and it is widely used in realizing on a large scale in neutral net
Handle Σ function.The dynamic range of collapsible Gilbert multipliers is big, the precision of multiplying is high.RBF neuron circuit moulds
Block includes two Gilbert multipliers, sqrt circuit and class Gaussian function generation circuit.Gilbert multipliers include
First to the 17th transistor M1~M17;The first transistor is to the 6th transistor M1~M6Emitter stage link together and connect high electricity
It is flat;The first transistor M1Base stage meet second transistor M2Base stage;The first transistor M1Colelctor electrode connect that to connect the 9th respectively brilliant
Body pipe M9Emitter stage and the tenth transistor M10Emitter stage;Second transistor M2Base stage connect the colelctor electrode of second transistor;
Second transistor M2Colelctor electrode meet the pipe M of the 7th crystal7Colelctor electrode;Third transistor M3Base stage meet the 4th transistor M4
Base stage;Third transistor M3Base stage meet third transistor M3Colelctor electrode;Third transistor M3Colelctor electrode connect the 8th crystal
Pipe M8Colelctor electrode;4th transistor M4Colelctor electrode pole meet the 11st transistor M respectively11Emitter stage and the tenth two-transistor
M12Emitter stage;5th transistor M5Base stage meet the 6th transistor M6Base stage;5th transistor M5Base stage connect the 5th crystal
Pipe M5Colelctor electrode;5th transistor M5Colelctor electrode meet the 14th transistor M14Colelctor electrode;6th transistor M6Colelctor electrode
The 17th transistor M is met respectively17Colelctor electrode and output Iout1;7th transistor M7Base stage meet Vw1;7th transistor M7's
Emitter stage connects the emitter stage and the 13rd transistor M of the 8th transistor respectively13Colelctor electrode;8th transistor M8Base stage connect
Vw2;9th transistor M9Base stage meet the tenth two-transistor M12Base stage;9th transistor M9Colelctor electrode connect the 11st respectively
Transistor M11Colelctor electrode and the 15th transistor M15Colelctor electrode;Tenth transistor M10Base stage meet the 11st transistor M11
Base stage;Tenth transistor M10Colelctor electrode meet the tenth two-transistor M respectively12Colelctor electrode and the 16th transistor M16Collection
Electrode;9th transistor M9With the tenth transistor M10Base stage meet V respectivelyinBoth positive and negative polarity;13rd transistor M13Base stage connect
Bias voltage Vbias;14th transistor M14The transistor M of base stage the 15th15Base stage;15th transistor M15Base stage
Connect its colelctor electrode;16th transistor M16Base stage meet the 17th pipe M17Base stage;16th transistor M16Base stage connect its current collection
Pole;The emitter stage of 13rd to the 17th transistor connects ground connection.The transistor size of Gilbert multipliers is referring to table 1.
Table 1
Pipe title | Breadth length ratio (W/L) | Pipe title | Breadth length ratio (W/L) |
M1 | 60/1 | M10 | 15/1 |
M2 | 60/1 | M11 | 15/1 |
M3 | 60/1 | M12 | 15/1 |
M4 | 60/1 | M13 | 100/1 |
M5 | 60/1 | M14 | 40/1 |
M6 | 60/1 | M15 | 40/1 |
M7 | 5/1 | M16 | 40/1 |
M8 | 5/1 | M17 | 40/1 |
M9 | 15/1 |
The transistor-level schematic of sqrt circuit is as shown in figure 4, its core is by M18, M19, M20And M21Structure
Into translinear structures, M20Pipe and M21The breadth length ratio of pipe is M18Pipe and M194 times of the breadth length ratio of pipe, the circuit can be real
Now to electric current sqrt.Specifically in the embodiment of the utility model one, the sqrt circuit includes the 18th to the
26 transistor M18~M26;The 18th transistor M18Colelctor electrode and the 19th transistor M19Base stage connection, which is used as, opens flat
Root circuits input Iin;18th transistor M18Base stage meets the 26th transistor M respectively26Base stage, the 19th transistor M19
Emitter stage and the 24th transistor M24Colelctor electrode;19th transistor M19Base stage meets the 21st transistor M21Base stage;Second
Ten transistor M20Base stage connects its colelctor electrode;21st transistor M21Emitter stage meets the 20th transistor M20Colelctor electrode;20th
One transistor M21Colelctor electrode meets the 20th two-transistor M22Colelctor electrode;20th two-transistor M22Base stage connects the 23rd crystal
Pipe base stage;19th transistor M19Colelctor electrode, the 20th two-transistor M22Emitter stage and the 23rd transistor M23Emitter stage connects
Be connected together high level;23rd transistor M23Colelctor electrode meets the 25th transistor M respectively25Colelctor electrode and output
Iout2;24th transistor M24Base stage meets the 25th transistor M respectively25Base stage and output end Vb;26th transistor
M26Colelctor electrode meets output Iout2;18th transistor M18Colelctor electrode and the 24th the 26th transistor M24~M26Emitter stage
Link together ground connection.The transistor size of sqrt circuit is referring to table 2.
Table 2
Pipe title | Breadth length ratio (W/L) |
M1 | 25/1 |
M2 | 25/1 |
M3 | 50/1 |
M4 | 50/1 |
M5 | 100/1 |
M6 | 100/1 |
M7 | 30/1 |
M8 | 30/1 |
M9 | 25/1 |
The transistor-level schematic of Gaussian circuit is as shown in figure 5, the class Gaussian function generation circuit includes the
27 to the 50th two-transistor M27~M52;27th transistor M27Emitter stage, the 28th transistor M28Emitter stage and
35th to the 40th two-transistor M35~M42Emitter stage, which links together, connects high level;27th transistor
M27Base stage meets the 27th transistor M respectively27Colelctor electrode and the 28th transistor M28Base stage;27th transistor M27
Colelctor electrode meet the 29th transistor M29Emitter stage;28th transistor M28Colelctor electrode meets the 30th transistor M30Hair
Emitter-base bandgap grading;29th transistor M29Base stage meets the 29th transistor M respectively29Colelctor electrode meets the 30th transistor M30Base stage;The
29 transistor M29Colelctor electrode meets the 31st transistor M31Colelctor electrode;30th transistor M30Colelctor electrode connects the 47th
Transistor M47Colelctor electrode;31st transistor M31Colelctor electrode connects its base stage;31st transistor M31Base stage connects respectively
31 transistor M31Colelctor electrode and the 30th two-transistor M32Base stage;31st transistor M31Emitter stage connects the 33rd
M33Colelctor electrode;30th two-transistor M32Colelctor electrode meets the 35th M35Colelctor electrode;30th two-transistor M32Emitter stage connects
34 transistor M34Colelctor electrode;33rd transistor M33Base stage meets the 33rd transistor M respectively33Colelctor electrode and the 3rd
14 transistor M34Base stage;33rd transistor M33Emitter stage, the 34th transistor M34Emitter stage, the 50th to the 5th
Ten two-transistor emitter stages link together ground connection;35th transistor M35Base stage connects the 35th transistor M respectively35
Colelctor electrode, the 36th transistor M36Base stage, the 41st transistor M41Base stage and the 40th two-transistor M42Base stage;3rd
16 transistor M36Colelctor electrode meets the 37th transistor M37Base stage;37th transistor M37Colelctor electrode connects the 40th respectively
Three transistor M43Colelctor electrode and meet the 45th transistor M45Colelctor electrode;37th transistor M37It is brilliant that base stage connects the 38th
Body pipe M38Base stage;38th transistor M38Base stage, the 39th transistor M39Colelctor electrode, the 40th two-transistor M42Current collection
Pole, which links together, meets output Iout3;39th transistor M39Base stage meets the 40th transistor M respectively40Base stage and the 41st
Transistor M41Colelctor electrode;40th transistor M40Colelctor electrode meets the 44th transistor M respectively44Colelctor electrode and the 46th crystalline substance
Body pipe M46Colelctor electrode;43rd transistor M43Base stage and the 46th transistor M46Base stage meets input V togetherin, the 43rd
Transistor M43Emitter stage meets the 44th transistor M respectively44Emitter stage and the 48th transistor M48Colelctor electrode;44th
Transistor M44Base stage meets V1;45th transistor M45Base stage meets V2;45th transistor M45Emitter stage connects the 40th respectively
Six transistor M46Emitter stage and the 49th transistor M49Colelctor electrode;47th transistor M47Base stage difference the 47th is brilliant
Body pipe M47Colelctor electrode, the 48th transistor M48Base stage, the 49th transistor M49Base stage;47th transistor M47Transmitting
Pole meets the 50th transistor M50Colelctor electrode;48th transistor M48Colelctor electrode meets the 51st transistor M51Colelctor electrode;4th
19 transistor M49Emitter stage meets the 50th two-transistor M52Colelctor electrode;50th transistor M50It is brilliant that base stage connects the 50th respectively
Body pipe M50Colelctor electrode, the 51st transistor M51Base stage, the 50th two-transistor M52Base stage.The circuit chip pipe size such as institute of table 3
Show.
Table 3
Pipe title | Breadth length ratio (W/L) | Pipe title | Breadth length ratio (W/L) | Pipe title | Breadth length ratio (W/L) |
M27 | 14/1 | M36 | 100/1 | M44 | 30/1 |
M28 | 14/1 | M37 | 100/1 | M45 | 30/1 |
M29 | 14/1 | M38 | 100/1 | M46 | 100/1 |
M30 | 14/1 | M39 | 100/1 | M47 | 100/1 |
M31 | 9/1 | M40 | 100/1 | M48 | 100/1 |
M32 | 9/1 | M41 | 100/1 | M49 | 100/1 |
M33 | 9/1 | M42 | 100/1 | M50 | 100/1 |
M34 | 9/1 | M43 | 30/1 | M51 | 100/1 |
M35 | 100/1 | M44 | 30/1 |
The method of work of above-mentioned RBF pattern classifier is:The data of point on two dimensional surface are by (Vx, Vy) input, the mould
Some points in plane are classified as Type1 by formula grader by judging, now network output IoutFor 1;And by other in plane
Point is classified as Type2, now network output IoutFor 0;The deterministic process of the wherein pattern classifier comprises the following steps:It is wherein defeated
Enter to hold VxAnd VyThe pattern classifier is entered data into, the function of input layer in RBF neural is completed;(Vx0, Vy0), (Vx0',
Vy0'), (Vx0", Vy0"), (Vx0" ', Vy0" ') center that represents four class Gaussian functions respectively is control input-hidden layer power
Value, (Vw1, Vw2), (Vw1', Vw2'), (Vw1", Vw2"), (Vw1" ', Vw2" '), (Vw1" ", Vw2" ") be respectively intended to control five it is hidden
Containing-output layer weights, (Vw1", Vw2") implicit-output layer weights of control are defaulted as 1, VbFor controlling to imply-output layer threshold
Value;First to fourth RBF neuron circuit modules realize Iout=bexp (- ((Vx-Vx0)2+(Vy-Vy0)2)/d) form of calculation,
B and d completes to imply layer function in RBF neural to be constant more than zero;First to the 5th Gilbert multipliers can be right
Hidden layer output is weighted summation operation, i.e. Iout=k1y1+k2y2+k3y3+k4y4+k5, k1, k2, k3, k4, k5For constant, y1,
Y2, y3, y4 are respectively output and the RBF neuron circuit modules of first, second, third, fourth RBF neuron circuit modules
Output, finally by output end IoutOutput category result, completes the function of output layer in RBF neural.
Gilbert multipliers include the first to the 17th transistor M1~M17, wherein the 7th transistor M7With the 8th crystal
Pipe M8With identical breadth length ratio, the 9th transistor M9, the tenth transistor M10, the 11st transistor M11With the 12nd M12With phase
Same breadth length ratio, transistor channel width is represented with W, L represents transistor channel length, CoxRepresent transistor unit area grid oxygen
Electric capacity, μpAnd μnThe channel mobility of hole and electronics is represented respectively, then the output current of the circuit is:
By Vw1With VinAnode be connected, use VxRepresent, by Vw2With Vin
Negative terminal be connected, use Vx0Represent, then can obtain:
That is Iout1=k (Vx-Vx0)2, wherein
The M of the sqrt circuit20Pipe and M21The breadth length ratio of pipe is M18Pipe and M194 times of the breadth length ratio of pipe, to input
Electric current sqrtI.e.
The class Gaussian function generation circuit includes the 27th to the 50th two-transistor M27~M52, it is assumed that the 30th
Six to the 41st transistor M36~M41With identical breadth length ratio, and it is the 35th transistor M35A times, the 40th
Two-transistor M42Breadth length ratio be the 35th transistor M35C times, the 48th transistor M48, the 49th transistor
M49, the 51st transistor M51, the 50th be transistor M52With identical breadth length ratio, and it is the 47th transistor M47、
50th transistor M50A times, obtaining output current is:
WhereinCoxFor transistor unit area grid oxygen electric capacity, μ moves for raceway groove
Shifting rate, W is transistor channel width, and L is transistor channel length, by the input voltage V for adjusting two differential pairs1And V2,
Adjust VinwcAnd Vw, so as to adjust the shape of the class Gaussian function of the circuit output;Required particular Gaussian function isSome discrete points of required particular Gaussian function are taken, by CADENCE software emulations, in formula (2)
Guidance under adjust circuit parameter it is final to obtain what is approached so that the corresponding point of class Gaussian function approaches these discrete points
Class Gaussian function Iout=bexp (- ((Vx-Vx0)2+(Vy-Vy0)2)/d)。
By the input voltage V for adjusting two differential pairs1And V2, V can be adjustedinwcAnd Vw, so as to adjust the circuit output
Class Gaussian function shape, as shown in Figure 6.Fig. 7 is b=40, the class that ideal Gaussian function during d=0.02 and adjustment are obtained
Gaussian function, with 0.01 is step-length in the range of -0.4~0.4, takes identical discrete point to two curves, respectively constitute vectorial A
And B, R is utilized in MATLAB softwares2=(A*B)2/A2*B2It can calculate and obtain the goodness of fit for R2=0.99775, therefore, should
Circuit simulation waveform can be fitted well with ideal Gaussian function.For required particular Gaussian function
(b and d are constant), we can take some discrete points, by CADENCE software emulations, be adjusted under the guidance of formula (2)
Circuit parameter, so that the corresponding point of class Gaussian function approaches these discrete points, finally obtains the class Gaussian function approached.Example
Such as, b=40, d=0.02 in the figure 7, class Gaussian function generation circuit parameter can be obtained for V by emulation1=-0.13V, V2=
0.13V,
In the specific embodiment of the utility model one, using MATLAB softwares and CADENCE softwares to the pattern based on RBF
Grader carries out the simulating, verifying of function.The circuit is completed based on 0.18 μm of CMOS technology parameter of SMIC, supply voltage
VddFor 1.8V, VssFor -1.8V.As shown in figure 8, in a graphic memory in (0,1), (1,0), (0,0), (1,0.9),
(0.9,1) five can not linear partition point (this five points can not be divided into (0,0) with straight line, (1,0.9),
(0.9,1) and (1,0), two groups of (0,1)), we can utilize four Gaussian function X1=G1(X, Y), X2=G2(X, Y), X3=
G3(X, Y) and X4=G4The effect of (X, Y) maps that to another four dimensional planes P (X1,X2,X3,X4) so that originally can not
Four points divided are in new plane P (X1,X2,X3,X4) interior redistribution, so as to realize linear partition.The utility model passes through solution
The pattern classification problems of certainly such a five points verifies the function of the pattern classifier, i.e., using the pattern classifier to two
The point inputted on dimensional plane is classified, and the data of the point on two dimensional surface are by (Vx, Vy) input, the pattern classifier is by plane
On point (1,0.9), (0.9,1) and (0,0) is classified as Type1, at this moment network output IoutFor 1;And by the point (1,0) in plane
(0,1) is classified as Type2, at this moment network output IoutFor 0.
Corresponding RBF neural is trained by MATLAB softwares, so as to obtain the parameter of neutral net, then
The parameter of neutral net is converted into the offset parameter needed for the pattern classifier, is loaded into circuit to realize specific pattern
Classification feature.Realize that different pattern classification functions need to write different training programs to be trained, in order that this practicality
The new pattern classification function of realizing five points, the training code write is as follows.The non-the utility model of programming content will
The content of protection.Neural Network Toolbox function net=newrb (P, T, eg, sc) in routine call MATLAB softwares comes
Neutral net is trained.In the training process, when the input of neutral net is vectorial for P, output vector constantly approaches the phase
Vector T is hoped, when the mean square deviation precision approached is less than eg=0.0001, neutral net deconditioning now obtains one
Variance precision eg<0.0001, dispersion constant sc=1 RBF neural shows training followed by 4 last orders
Neural network parameter afterwards.
RBF neural trains code:
Emulated by MATLAB, obtained neural network parameter as shown in table 4.
Table 4
The parameter of neutral net obtained by being trained by MATLAB is converted into inclined needed for the paralleling model grader
Parameter is put, input 0 is represented with 0V, input 1 is represented with 0.3V, then can obtain required circuit parameter values by calculating, its
Middle process emulation can obtain V1And V2Respectively 0.13V and -0.13V, it is determined that the shape of class Gaussian function.Required parameter is direct
It is loaded on the paralleling model grader circuit, is emulated with CADENCE, (wherein 0V is represented its simulation result as shown in Figure 9
0,0.3V of input represents input 1, and 0.27V represents input 0.9).From the simulation result can be seen that when input for point (1,
0.9), when (0.9,1) or (0,0), it is output as 1;When input is point (1,0) or (0,1), 0 is output as, correctly realizing will
Five click through row mode classification, and demonstrating the pattern classifier has good pattern classification function.
Above is preferred embodiment of the present utility model, all changes made according to technical solutions of the utility model are produced
Function without departing from technical solutions of the utility model scope when, belong to protection domain of the present utility model.
Claims (4)
1. a kind of pattern classifier based on RBF, it is characterised in that:Including first to fourth RBF neuron circuits module, first
To the 4th resistance and the first to the 5th Gilbert multipliers;
The current output terminal of described first to the 5th Gilbert multipliers is sequentially connected the output end I as graderout;First
The output of RBF neuron circuits connects the first input of first resistor one end and the first Gilbert multipliers respectively;First resistor is another
One end is grounded respectively and the first Gilbert multipliers second are inputted;The output of 2nd RBF neuron circuits connects second resistance respectively
One end and the first input of the 2nd Gilbert multipliers;The second resistance other end is grounded and the 2nd Gilbert multipliers respectively
Two inputs;The output of 3rd RBF neuron circuits connects 3rd resistor one end respectively and the 4th Gilbert multipliers first are inputted;
The 3rd resistor other end is grounded respectively and the 4th Gilbert multipliers second are inputted;The output difference of 4th RBF neuron circuits
Connect the 4th resistance one end and the 5th Gilbert multipliers;The 4th resistance other end is grounded and the 5th Gilbert multipliers respectively
Second input;
First output end of the first RBF neuron circuit modules is Vx, the second output end is Vx0, the 3rd output end is Vy, the
Four output ends are Vy0;First output end of the 2nd RBF neuron circuit modules is Vx, the second output end is Vx0', the 3rd is defeated
It is V to go out endy, the 4th output end is Vy0';First output end of the 3rd RBF neuron circuit modules is Vx, the second output end
For Vx0' ', the 3rd output end is Vy, the 4th output end is Vy0'';First output end of the 4th RBF neuron circuit modules
For Vx, the second output end is Vx0' ' ', the 3rd output end is Vy, the 4th output end is Vy0''';First Gilbert multipliers
One output end is Vw1, the second output end is Vw2, the 2nd the first output end of Gilbert multipliers is Vw1', the second output end is
Vw2';3rd the first output end of Gilbert multipliers is Vw1' ', the second output end is Vw2'';4th Gilbert multipliers
One output end is Vw1' ' ', the second output end is Vw2''';5th the first output end of Gilbert multipliers is Vw1' ' ' ', second
Output end Vw2' ' ' ', the 3rd end of the 3rd Gilbert multipliers is Vb。
2. the pattern classifier according to claim 1 based on RBF, it is characterised in that:Gilbert multipliers include
First to the 17th transistor M1~M17;The first transistor is to the 6th transistor M1~M6Emitter stage link together and connect high level;
The first transistor M1Base stage meet second transistor M2Base stage;The first transistor M1Colelctor electrode connect and connect the 9th transistor respectively
M9Emitter stage and the tenth transistor M10Emitter stage;Second transistor M2Base stage connect the colelctor electrode of second transistor;Second
Transistor M2Colelctor electrode meet the pipe M of the 7th crystal7Colelctor electrode;Third transistor M3Base stage meet the 4th transistor M4Base
Pole;Third transistor M3Base stage meet third transistor M3Colelctor electrode;Third transistor M3Colelctor electrode meet the 8th transistor M8
Colelctor electrode;4th transistor M4Colelctor electrode pole meet the 11st transistor M respectively11Emitter stage and the tenth two-transistor M12's
Emitter stage;5th transistor M5Base stage meet the 6th transistor M6Base stage;5th transistor M5Base stage meet the 5th transistor M5
Colelctor electrode;5th transistor M5Colelctor electrode meet the 14th transistor M14Colelctor electrode;6th transistor M6Colelctor electrode point
The 17th transistor M is not met17Colelctor electrode and output Iout1;7th transistor M7Base stage meet Vw1;7th transistor M7Hair
Emitter-base bandgap grading connects the emitter stage and the 13rd transistor M of the 8th transistor respectively13Colelctor electrode;8th transistor M8Base stage meet Vw2;
9th transistor M9Base stage meet the tenth two-transistor M12Base stage;9th transistor M9Colelctor electrode connect the 11st crystal respectively
Pipe M11Colelctor electrode and the 15th transistor M15Colelctor electrode;Tenth transistor M10Base stage meet the 11st transistor M11Base
Pole;Tenth transistor M10Colelctor electrode meet the tenth two-transistor M respectively12Colelctor electrode and the 16th transistor M16Colelctor electrode;
9th transistor M9With the tenth transistor M10Base stage meet V respectivelyinBoth positive and negative polarity;13rd transistor M13Base stage connect biasing
Voltage Vbias;14th transistor M14The transistor M of base stage the 15th15Base stage;15th transistor M15Base stage connect its collection
Electrode;16th transistor M16Base stage meet the 17th pipe M17Base stage;16th transistor M16Base stage connect its colelctor electrode;The
The emitter stage of 13 to the 17th transistors connects ground connection.
3. the pattern classifier according to claim 1 based on RBF, it is characterised in that:RBF neuron circuit modules include
Two Gilbert multipliers, sqrt circuit and class Gaussian function generation circuit;
Sqrt circuit includes the 18th to the 26th transistor M18~M26;18th transistor M18Colelctor electrode and the 19th
Transistor M19Base stage connection is used as sqrt circuit input end Iin;18th transistor M18It is brilliant that base stage connects the 26th respectively
Body pipe M26Base stage, the 19th transistor M19Emitter stage and the 24th transistor M24Colelctor electrode;19th transistor M19Base stage connects
21st transistor M21Base stage;20th transistor M20Base stage connects its colelctor electrode;21st transistor M21Emitter stage connects
20 transistor M20Colelctor electrode;21st transistor M21Colelctor electrode meets the 20th two-transistor M22Colelctor electrode;22nd is brilliant
Body pipe M22Base stage connects the 23rd transistor base;19th transistor M19Colelctor electrode, the 20th two-transistor M22Emitter stage and
23rd transistor M23Emitter stage, which links together, connects high level;23rd transistor M23Colelctor electrode connects the 20th respectively
Five transistor M25Colelctor electrode and output Iout2;24th transistor M24Base stage meets the 25th transistor M respectively25Base stage with
Output end Vb;26th transistor M26Colelctor electrode meets output Iout2;18th transistor M18Colelctor electrode and the two ten four the second
16 transistor M24~M26Emitter stage links together ground connection.
4. the pattern classifier according to claim 3 based on RBF, it is characterised in that:The class Gaussian function produces electricity
Road includes the 27th to the 50th two-transistor M27~M52;27th transistor M27Emitter stage, the 28th transistor M28
Emitter stage and the 35th to the 40th two-transistor M35~M42Emitter stage, which links together, connects high level;Described 27th
Transistor M27Base stage meets the 27th transistor M respectively27Colelctor electrode and the 28th transistor M28Base stage;27th is brilliant
Body pipe M27Colelctor electrode meet the 29th transistor M29Emitter stage;28th transistor M28Colelctor electrode connects the 30th crystal
Pipe M30Emitter stage;29th transistor M29Base stage meets the 29th transistor M respectively29Colelctor electrode meets the 30th transistor M30
Base stage;29th transistor M29Colelctor electrode meets the 31st transistor M31Colelctor electrode;30th transistor M30Colelctor electrode connects
47 transistor M47Colelctor electrode;31st transistor M31Colelctor electrode connects its base stage;31st transistor M31Base stage is distinguished
Meet the 31st transistor M31Colelctor electrode and the 30th two-transistor M32Base stage;31st transistor M31Emitter stage connects the 3rd
13 M33Colelctor electrode;30th two-transistor M32Colelctor electrode meets the 35th M35Colelctor electrode;30th two-transistor M32Emitter stage
Meet the 34th transistor M34Colelctor electrode;33rd transistor M33Base stage meets the 33rd transistor M respectively33Colelctor electrode and
34th transistor M34Base stage;33rd transistor M33Emitter stage, the 34th transistor M34Emitter stage, the 50th to
50th two-transistor emitter stage links together ground connection;35th transistor M35Base stage connects the 35th transistor respectively
M35Colelctor electrode, the 36th transistor M36Base stage, the 41st transistor M41Base stage and the 40th two-transistor M42Base stage;The
36 transistor M36Colelctor electrode meets the 37th transistor M37Base stage;37th transistor M37Colelctor electrode connects the 4th respectively
13 transistor M43Colelctor electrode and meet the 45th transistor M45Colelctor electrode;37th transistor M37Base stage connects the 38th
Transistor M38Base stage;38th transistor M38Base stage, the 39th transistor M39Colelctor electrode, the 40th two-transistor M42Collection
Electrode, which links together, meets output Iout3;39th transistor M39Base stage meets the 40th transistor M respectively40Base stage and the 40th
One transistor M41Colelctor electrode;40th transistor M40Colelctor electrode meets the 44th transistor M respectively44Colelctor electrode and the 46th
Transistor M46Colelctor electrode;43rd transistor M43Base stage and the 46th transistor M46Base stage meets input V togetherin, the 40th
Three transistor M43Emitter stage meets the 44th transistor M respectively44Emitter stage and the 48th transistor M48Colelctor electrode;40th
Four transistor M44Base stage meets V1;45th transistor M45Base stage meets V2;45th transistor M45Emitter stage connects the 4th respectively
16 transistor M46Emitter stage and the 49th transistor M49Colelctor electrode;47th transistor M47Base stage difference the 47th
Transistor M47Colelctor electrode, the 48th transistor M48Base stage, the 49th transistor M49Base stage;47th transistor M47Hair
Emitter-base bandgap grading meets the 50th transistor M50Colelctor electrode;48th transistor M48Colelctor electrode meets the 51st transistor M51Colelctor electrode;The
49 transistor M49Emitter stage meets the 50th two-transistor M52Colelctor electrode;50th transistor M50Base stage connects the 50th respectively
Transistor M50Colelctor electrode, the 51st transistor M51Base stage, the 50th two-transistor M52Base stage.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2016205361918 | 2016-06-03 | ||
CN201620536191 | 2016-06-03 |
Publications (1)
Publication Number | Publication Date |
---|---|
CN206480025U true CN206480025U (en) | 2017-09-08 |
Family
ID=59755809
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201620745796.8U Expired - Fee Related CN206480025U (en) | 2016-06-03 | 2016-07-15 | Pattern classifier based on RBF |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN206480025U (en) |
-
2016
- 2016-07-15 CN CN201620745796.8U patent/CN206480025U/en not_active Expired - Fee Related
Similar Documents
Publication | Publication Date | Title |
---|---|---|
PETRŽELA et al. | Modeling Deterministic Chaos Using Electronic Circuits. | |
Duan et al. | Stability and l1-gain analysis for positive 2D T–S fuzzy state-delayed systems in the second FM model | |
CN106897821A (en) | A kind of transient state assesses feature selection approach and device | |
CN104598766B (en) | Towards the space reasoning by cases method of Comprehensive Evaluation of Reservoirs | |
CN114841031A (en) | Method for calculating diffusion concentration of simulated toxic and harmful gas in three-dimensional virtual training environment | |
Zhang et al. | A power-type varying gain discrete-time recurrent neural network for solving time-varying linear system | |
CN105976024B (en) | Pattern classifier and its working method based on RBF | |
CN105045948B (en) | A kind of transmitter guard box configuration diagram automatic batch generation method | |
CN206480025U (en) | Pattern classifier based on RBF | |
CN205788258U (en) | Function approximator based on RBF | |
CN106067063B (en) | RBF neuron circuit and its working method | |
Waibel et al. | Physics meets machine learning: Coupling FFD with regression models for wind pressure prediction on high-rise facades | |
CN206003142U (en) | RBF neuron circuit | |
Ding et al. | Finite-frequency model reduction of discrete-time T–S fuzzy state-delay systems | |
CN106067064A (en) | Function approximator system based on RBF | |
CN103136430B (en) | Based on the solid propellant formula optimization of genetic algorithm and energy feature graph | |
He et al. | Research on the teaching mode of university virtual laboratory based on component technology | |
CN109523143B (en) | Land evaluation method based on multi-granularity calculation | |
Hou et al. | A novel modeling technique for efficiently computing 3-D capacitances of VLSI multilevel interconnections-BFEM | |
Sunori et al. | K-Means and Competitive Neural Network Based Clustering of Climate Data | |
CN108681502A (en) | A kind of CPS software energy consumption computational methods based on hierarchic parallel algorithm | |
Venkataswamy et al. | AMSDAT: Integrated analog and mixed-signal design optimization framework for SoC applications | |
CN107609274A (en) | Two-dimentional magnetostatic field parallel finite element method method based on transmission line Yu rank scheduling method | |
CN103324705B (en) | Extensive vector field data processing method | |
Panopio et al. | Matrix Methods for Enhanced DC Analysis: A Python-Based Approach |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170908 Termination date: 20190715 |