The priority of the United States Non-Provisional application number 13/952,402 that patent application claims were submitted on July 26th, 2013,
The priority of the United States Non-Provisional application number 13/791,201 that patent application claims were submitted on March 8th, 2013 and be its portion
Divide and continue, this two applications are incorporated herein by reference, for various purposes.
Specific embodiment
The present invention is directed to data communication system and method.More specifically, various implementation methods of the invention provide communication
Interface, the communication interface is configured to optical communication networks and uses nDSQ forms to pass through high bandwidth transmission data.At certain
In a little implementation methods, the various devices (for example, spine switches and leaf interchanger) in trunk-leaf network framework use communication
Interface, so allows to share mass data between servers.
In the past few decades, with cloud computing and the appearance of data center, the demand to the webserver has changed.Example
Such as, it is no longer appropriate or suitable by the three-level configuration of long-term use, because the application program of distribution needs more flat network
Framework, wherein, server virtualization allows server operation repetitive.For example, multiple servers can be provided commonly for performing request
Task.For the multiple servers for wanting parallel running, these servers generally need to be quickly shared bulk information in period, with
Multitiered network framework (for example, network switch etc.) is run data through to move back and forth conversely.
There is provided leaf-backbone-type the network architecture, with preferably allow server it is parallel running and between servers
Quick mobile data, there is provided high bandwidth and low latency.Generally, leaf-core network framework uses frame topcross, and the interchanger can
With one group of spine switches of non-obstruction of direct access server node and Opposite direction connection, these spine switches have enough
Bandwidth, to allow server cluster to be connected to each other and share mass data.
In nowadays typical leaf-core network, kilomegabit data are shared between servers.In some network architectures
In, the webserver in same levels has some peer links for data sharing.Regrettably, it is such
The bandwidth of setting is generally inadequate.It is to be understood that embodiments of the present invention are encoded in leaf-backbone architectures using nDSQ, should
Framework allows to transmit a large amount of (in trunk grade, up to terabit data) data by optic network.
Following description is proposed, can allow for those skilled in the art's construction and use the present invention, and specific
The present invention is combined under the background of application.For those skilled in the art, various modifications and a large amount of use in different application
Way is it is clear that and during the General Principle that limits herein can apply to a large amount of implementation methods.Therefore, the present invention not
It is restricted to proposed implementation method, but meets consistent with principle disclosed herein and novel feature widest
Scope.
In the following detailed description, multiple details are set forth, to provide more thorough understanding of the invention.However,
, it will be apparent to those skilled in the art that the present invention can be put into practice, it is not necessary to be limited to these details.In other cases, use
The form (rather than in detail) of block diagram shows well-known construction and device, in case the present invention is hard to understand.
Reader notes, is submitted to simultaneously with this specification and the whole papers for public inspection are disclosed together with this specification
And document, and the content of all these papers and document is incorporated herein by reference.In this specification (including any appended power
Profit require, summary and diagram) disclosed in all features can be by the replacement feature generation for identical, equivalent or similar purpose
Replace, unless otherwise expressly provided.Therefore, unless otherwise expressly provided, otherwise it is disclosed each be characterized in the equivalent of universal serial
Or the only one example in similar features.
Further, " device " for performing specific function or the " step for performing specific function are not yet explicitly specified
Suddenly any element in the claims " is not interpreted as specified in the 6th section of United States Code No. 35 the 112nd section " device "
Or " step " clause.In particular, " step includes (step of) " is used in claim herein or " behavior includes
(act of) " is not intended to use the regulation of the 6th section of United States Code No. 35 the 112nd section.
It should be noted that if you are using, then be marked with left, right, front and rear, top, bottom, forward, the back side, up time
Pin and convenient purpose is used only for counterclockwise, it is not intended that represent any specific fixed-direction.Conversely, these labels are used
Relative position and/or direction between the different piece for being reflected in target.
Fig. 1 is the simplification figure of the leaf-backbone architectures 100 for showing an implementation method of the invention.This diagram is only
Only it is an example, the example should not excessively limit the scope of claim.Those skilled in the art will appreciate that many changes
Change, replace and change.Leaf-backbone architectures 100 include server 120, leaf interchanger 110 and spine switches 103.Manage
Solution, as needed and concrete application, the quantity of server and interchanger and sets and can change.As shown in fig. 1, each
Server can be connected to more than one leaf interchanger.For example, server 121 is connected to leaf interchanger 111 and 112.Equally, take
Business device 122 is connected to leaf interchanger 111 and 112, and server 123 is also such.In an illustrative embodiments, server
121 optical communications links mapped by using pulse amplitude modulation (PAM) and nDSQ are connected to leaf interchanger 111.In this hair
In bright various implementation methods, nDSQ and PAM2, PAM4, PAM8, PAM 12, PAM 16 and/or PAM other changes can be with
It is combined with optical communications link and is used.The bandwidth of the optical communications link between server 121 and leaf interchanger 111 can be with
More than 10 gigabit/second.Each leaf interchanger (for example, leaf interchanger 111) can be connected to 10 or more servers.
In one implementation, leaf interchanger has at least 100 gigabit/second of bandwidth.
In a detailed embodiment, leaf interchanger includes acceptor device, is configured as receiving 4 communication channels,
And each channel input datas and can be configured as PAM-2 forms by 25 gigabit/second of transmission.For example, can lead to
PCI-e interfaces are crossed, input data is received from processor.Equally, server (for example, server 121) includes communication interface, institute
Communication interface is stated to be configured to 100 gigabit/second of transmission and receive (for example, individual channel has the 4 of 25 gigabit/second
Individual channel), and it is compatible with the communication interface of leaf interchanger.Spine switches equally include being used for PAM format transmissions and connecing
Receive the communication interface of data.Spine switches can have mass communication channel, to accommodate a large amount of leaf interchangers, each interchanger
The switching of a large amount of servers is provided.
Leaf interchanger is connected to spine switches.As shown in fig. 1, each leaf interchanger is connected to spine switches 101
With 102.For example, leaf interchanger 111 is connected to spine switches 101 and 102, and leaf interchanger 113 and 114 is also such.
In a particular implementation, each spine switches is configured with the billion bps of bandwidth 3.2, and this bandwidth is sufficiently large, to pass
32 optical communications links are sent, each link has 100 gigabit/second.According to specific implementation, additionally it is possible to other configurations
And bandwidth.
By the construction 100 for showing in Fig. 1, server can effectively be communicated with one another by high bandwidth.In server
Optical communications link is used between leaf interchanger and also between leaf interchanger and spine switches, and PAM is used for light
Learn network service.
It is to be understood that above-mentioned PAM communication interfaces can be realized according to nowadays communication standard form factor.Additionally, with biography
Bulk cargo is put and is compared, and the network transceivers according to the embodiment of the present invention provided by high efficiency levels can have much lower
Power consumption and smaller form factor.Fig. 2 is the formal cause of the communicator for showing an implementation method of the invention
The simplification figure of son.Nowadays, the pluggable interfaces of C (CFP) standard is widely used in giga-bit network system.Electrically connected based on traditional
The CFP transceivers for connecing realize high bandwidth usually using 10x10 gigabit/second of circuit.Connected by optics, CFP transceivers can
To use 10x10 gigabit/second of configuration, 4x25 gigabit/second of configuration or other configurations.It is to be understood that by using optics
Communication link and PAM forms, transceiver of the invention can have the form more much smaller than shown CFP and CFP2
The factor.In various embodiments, communication interface of the invention can have form factor CFP4 or QSFP, its size ratio
CFP is much smaller.In addition to smaller form factor, the power consumption of communication interface of the invention can much more
It is small.In a detailed embodiment, by the form factor of QSFP, power consumption can as little as 3W, this is about with CFP forms
The 1/4 of the conventional transceiver of the factor.Lower level power consumption helps disposing thousands of (sometimes millions of) such communications
The data center of device saves energy.
Fig. 3 A are the simplification figures of the communication interface 300 for showing an implementation method of the invention.This diagram is only
It is an example, the example should not excessively limit the scope of claim.Those skilled in the art will appreciate that many changes,
Replace and change.Communication interface 300 includes transmitter module 310 and receiver module 320.Transmitter module 310 includes
Receiver 311, encoder 312 and PAM modulating drivers 313.
In one embodiment, communication interface 300 is configured to 4 channels and receives input data, wherein, each
Channel configuration is 25 gigabit/second and is configured as PAM-2 forms.Using transmitter module 310, modulator 316 and swash
Light device 314, communication interface 300 is processed by 25 gigabit/second of data of reception from each in 4 input channels,
And the optical data stream modulated by 100 gigabit/second of bandwidth for transmission PAM.It should be understood that, additionally it is possible to other bands
Width, for example, 40Gbps, 400Gbps and/or other bandwidth.
As illustrated, transmitter module 310 receives 4 data channels.It is to be understood that in addition to PAM-2 forms, also
Other deformations (for example, PAM4, PAM8, PAM12, PAM 16 etc.) of pulse amplitude modulation can be used.Transmitter module 310 is wrapped
Functional block 311 is included, the functional block includes clock data recovery circuit, and it is configured as receiving described defeated from 4 communication channels
Enter data.In various embodiments, functional block 311 further includes multiplexer, for combining 4 data channels.Example
Such as, 4 shown data of passage come from PCI-e interfaces 350.For example, interface 350 is connected to one or more processors.
In a detailed embodiment, two 2 are used in functional block 311:1 multiplexer.For example, being received from 4 passages
Data be high-speed data-flow without clock signal.Additionally, receiver 311 includes being associated with predetermined frequency reference
Clock signal.In various embodiments, receiver 311 is configured with phaselocked loop (PLL) and received data pair
It is accurate.
Transmitter module 310 further includes encoder 312.As shown in Figure 3, encoder 312 includes forward error correction
(FEC) encoder.Additionally, encoder 312 provides error detection and/or correction as needed.For example, as described above, in PAM-2
Data are received in form.Received data includes that redundancy (for example, one or more redundant digits) helps encoder 312 to detect mistake
By mistake.In a detailed embodiment, low-density checksum (LDPC) code is used.Encoder 312 is configured as from showing
The data encoding received in 4 channels for showing, can be by the kilomegabit of bandwidth 100 (for example, having 25 kilomegabits/number with generation
According to combination 4 channels) by optical communications link transmit data flow.For example, each data for receiving has PAM-2 lattice
Formula, and encoded data stream is 4 combinations of data channel and with nDSQ forms.As described below in more detail, by number
According to stream encryption into nDSQ symbols, and n is the quantity of dimension.For example, n dimension constellations are used to map the data into nDSQ symbols.
PAM modulating drivers 313 are configured as driving the driving data encoded by encoder 312.In various implementation methods
In, receiver 311, encoder 312 and modulating driver 313 are integrated and as a part for transmitter module 310.
PAM modulators 316 are configured as the data of modulation transmissions device module 310, and use laser 314 by reception
Electric signal is converted into optical signal for example, modulator 316 generates optical signal by the transmission rate of 100 kilomegabit per second.It is appreciated that
Be, additionally it is possible to other speed, for example, 40Gbps, 400Gbps etc..With PAM forms (for example, PAM-8 forms, PAM12,
PAM16 etc.) transmission optical signal.In various embodiments, laser 314 includes distributed Feedback (DFB) laser.According to should
With, can also use other kinds of laser technique, for example, vertical cavity surface emitting laser (VCSEL) etc..
Fig. 3 B are the simplification figures of the segmented optical modulator for showing an implementation method of the invention.This diagram is only
Only it is an example, the example should not excessively limit the scope of claim.Those skilled in the art will appreciate that many changes
Change, replace and change.For example, the PAM signal modulations of modulation are transmitted into by optical communications link.
Now, return and see Fig. 3 A.Communication interface 300 is configured to receive and transmit signal.Receiver module 320 includes light
Detector 321, its input data signal of conversion with format optical, converts optical signals into electric signal.In various embodiment party
In formula, photo-detector 321 includes indium gallium arsenide material.For example, photo-detector 321 can be the pole of photoelectricity two based on semiconductor
Pipe, for example, p-n photodiodes, p-i-n photodiodes, avalanche photodide.Photo-detector 321 and the coupling of amplifier 322
Close.In various embodiments, amplifier includes linear transimpedance amplifier (TIA).It is to be understood that using TIA, can be with vacation
If remote multi-mode (LRM) has high bandwidth (for example, 100Gb/s or even more big).For example, TIA helps use electrically dispersion
Compensation (EDC) compensates optical dispersive in electrical domain.In some embodiments, amplifier 322 also includes limiting amplifier.Put
Big device 322 is used to produce the signal with electrical domain from input optical signal.In some embodiments, transmission data it
Before, further signal transacting can also be applied, for example, recovering (CDR) by the data clock that phaselocked loop is performed.
Amplification data signal from amplifier 322 is processed by analog-digital converter (ADC) 323.In specific embodiment
In, ADC 323 can be baud rate ADC.For example, ADC is configured as the signal of amplification being converted into data signal, numeral letter
Number it is formatted as the signal of 100 kilomegabit per second with PAM forms.For example, data-signal can have represents data flow
NDSQ symbols.Functional block 324 is configured as processing 100Gb/s data flows, and is encoded into single 4 with 25Gb/s
Stream.For example, the optical data stream of the input received by photo-detector 321 is with the 100Gb/s bandwidth of PAM-8 forms, and
In square frame 324, with 25Gb/s bandwidth 4 data flows of the generation with PAM-2 forms.This 4 data flows pass through 4 with 25Gb/s
Communication channel is transmitted by transmitter 325.
It is to be understood that the implementation method for describing in figure 3 can have various changes.For example, according to application (for example,
Server, leaf interchanger, spine switches etc.), the channel (for example, 4,8,16 etc.) of varying number and difference can also be used
Bandwidth (for example, 10Gb/s, 40Gb/s, 100Gb/s, 400Gb/s, 3.2Tb/s etc.).
Fig. 4 is the simplified block diagram for showing coding and mapping treatment according to the embodiment of the present invention.This diagram is only
Only it is an example, the example should not excessively limit the scope of claim.Those skilled in the art will appreciate that many changes
Change, replace and change.Used as an example, in figure 3 a, encoder 401 and nDSQ mappers 402 are the two of FEC 312
Individual part.In a specific embodiment, encoder 401 is configured as the position of data flow being divided into highest significant position (MSB) and most
Low order (LSB).NDSQ mappers 402 are supplied to using MSB as data block.LSB is encoded with various encoding schemes.For example,
BCH and/or RS encoding schemes can be used, LSB is encoded.
It is to be understood that the n that data flow is mapped to for transmitting is tieed up into both sides quadrature amplitude modulation (nDSQ) encoded signal
In symbol.It is to be understood that being carried out data transmission using nDSQ symbols, it is possible to achieve high transfer rate.In various embodiment party
In formula, the invention provides being encoded for nDSQ there is provided high efficiency and the technology of pinpoint accuracy.Fig. 5 A and 5B are to show root
According to the simplification figure for the constellation in DSQ mappings of an embodiment of the invention.This diagram is only an example, should
Example should not excessively limit the scope of claim.Those skilled in the art will appreciate that many changes, replace and change.
For example, a2kRepresent the symbol in data flow, and a2k+1Represent the next symbol in data flow.As shown in Figure 5 A,
Closely packed together there is provided the DSQ32 codings of 2 dimensions, and data symbol.
Fig. 5 B show that DSQ32 is encoded, wherein, eliminate the data symbol of half.A number is removed by every one
According to symbol, increase from a data symbol to the distance of its adjacent symbol, this produces 3dB gains in signal to noise ratio (SNR).Example
Such as, the change of transmission rate is speed=2.5/symbol.By will be from DSQ constellations by 5 bit mappings to 2 PAM8 symbols
In, it is possible to achieve the reduction of symbol.It is to be understood that the reduction of adjacent-symbol can be performed on n dimension DSQ constellations.For example, right
In three-dimensional DSQ constellations, wherein, axle is 3k、3k+1、3k+2, character density can equally reduce.Although data symbol subtracts in a constellation
It is few, message transmission rate can be reduced, but improve SNR.
Used as an example, DSQ32 can be construed to (6,5) code, wherein, provide odd even side by following equation 1
Journey:
a2k+a2k-1=0mod2 (equation 1)
PAM8 symbols can be decomposed according to equation 2 below:
(equation 2)
According to definition, the nD-DSQ symbol definitions of dimension dim are:
As demonstrated, n dimensions DSQ codings include 3*dim-1 bit mappings to dim PAM8 symbols (as nD-DSQ symbols)
In.For example, nD-DSQ codings are equal to the 1 dislocation error detection on dim-1LSB (PAM2) position.Ratio is ratio=3-1/dim
Position/symbol.SNR gain (each is adjacent) is still 3dB, but adjacent quantity with dim increases (described later on).
Fig. 6 is the simplification figure for showing PAM mappings according to the embodiment of the present invention.This diagram is only a reality
Example, the example should not excessively limit the scope of claim.Those skilled in the art will appreciate that many changes, replace and
Modification.As shown in Figure 6, input bit is divided into MSB and LSB.For MSB, (M-1) * dim numbers of position are mapped by PAM- (M-1)
Device maps, to provide dim PAM- (M-1) symbol.For example, the MSB being mapped is { 0,2,4,6 ... M-2 } form.For
LSB, (dim-1) individual number of position is processed by increasing parity check bit, to generate dim number of position.For example, odd even school
Test position for remove LSB " odd number " and, so that the even-odd check of DSQ symbols is even number all the time.PAM2 is mapped for LSB bit,
And increase parity check bit for error detection purpose.PAM2 and PAM- (M-1) outputs are added together, to provide n dimensions DSQ
Symbol.
Fig. 7 is the simplification figure for showing PAM8 mapping treatment according to the embodiment of the present invention.This diagram is only one
Individual example, the example should not excessively limit the scope of claim.Those skilled in the art will appreciate that many changes, replace
And modification.As shown in Figure 7, MSB is mapped with PAM4 gray mappings.LSB is mapped with PAM2 mappings, and even-odd check
Number for promoting the 1 of LSB is even number.Parity check bit is used to detect every dim single-bit error, and application is single wrong
Cause delay part maximum likelihood correction.It is coded of MSB and LSB is combined, this is nature multilevel code structure.For example, closest
The neighbours of given nD-DSQ constellation points are corresponding with 2 of the upset in the dim of LSB.Compared with uncoded situation, this
Corresponding to 3dB Euclidean distance gains.For example, each nD-DSQ symbol shows the neighbour of the par limited by following formula
Occupy:
3xnchoosek(dim, 2)
For example, it is contemplated that the LSB bit of correction, MSB shows 6dB gains.As dimension n increases, symbol reduces what is formed
Ratio punishment is than reducing.
As described above, before being mapped in DSQ symbols, MSB and LSB is encoded first.Fig. 8 is shown according to the present invention
An implementation method have coding MSB and LSB bit PAM8 map technique simplification figure.This diagram is only a reality
Example, the example should not excessively limit the scope of claim.Those skilled in the art will appreciate that many changes, replace and
Modification.As shown in Figure 8, LSB is encoded with ratio R 0.For example, LSB is encoded with BCH and/or RS codes.Using stronger
Coding is with ratio R 0 by LSB channel codings.MSB channels are encoded using rate code R1 higher.In order to have effect higher
The coding of rate, LSB and DSQ is optimised, and this depends on the property of possible error event.For example, for code sign, can be used non-
Binary coding is encoded to LSB, forms the multiple of (dim-1) position.In a specific embodiment, (dim-1) intertexture two is used to enter
Code processed is encoded to LSB.For example, (dim-1) Bose-Chaudhuri-Hocquenghem Code for interweaving to be supplied to the PAM2 mappers for LSB.Specific real
Apply in mode, the BCH (n, k, t) that (dim-1) interweaves is operated with GF (2^m) codings of parallel running.For example, unless channel is made
Into the error burst longer than dim, BCH decoders are displayed in the single-bit error of the independent random of its input.Ratio is ratio
Rate=2R1+R0* (dim-1)/dim position/symbol.Optimized relation between BCH code and dim so that R0* (dim-1)/
Dim increases and meets deferred constraint as far as possible.For example, the baud rate of coding can be 100e9*257/256/Rate.BCH blocks
It is n*dim/Baud_Rate to postpone (BL).MSB external codes contribute to total delay, treatment is only helped in decoder and is postponed.
Fig. 9 A and 9B are the simplification figures of the decoding technique of the nDSQ symbols for showing an implementation method of the invention.This
Individual diagram is only an example, and the example should not excessively limit the scope of claim.Those skilled in the art will appreciate that
To many changes, replace and modification.In figure 9 a, by n dimension DSQ symbol decodings.Symbol is received by channel 901.Passing through
During the transmission technique of channel 901, additive white Gaussian noise (AWGN) can be introduced, wherein, SNR is SNR=Es/sigma2.Such as
Shown in figure, DSQ symbols are split by PAM food slicers, wherein, MSB (for example, 2 MSB) is processed on top section, LSB is in bottom
It is upper to process and decoded by Wagner decoders.In one example, Wagner decoders are used to extract Soft Inform ation.It is to be appreciated that
, other kinds of decoder can also be used.For example, when LSB is processed for error checking, using even-odd check
Position.
For example, error generator and PAM8 food slicers are a parts for DSP core.In various embodiments, mistake letter
Number used by all self-adaption loops.The output that xor logic is used for the Wagner decoders according to determines whether have
0.If wrong, then xor logic device determines whether it is " 1 ".
Fig. 9 B show nDSQ decoders, wherein, MSB is processed by PAM4 food slicers.For example, LSB is used as multiplexer
951 input, is used to determine whether to increase " 1 ".Specified by below equation nD-DSQ (dim) symbol error rate (SER) it is near
Like value:
Factor sqrt (2) in Q functions represents 3dB SNR gains, and multiplicity factor 3nchoosek(dim, 2) is
The average number of " 3dB " neighbours of each nD-DSQ symbol.As dim increases, ratio=3-1/dim " has with lower
Effect " SNR gain is reduced for cost.According to application, Dominant error events are the double bit errors of each nD-DSQ symbol.For example,
In 20dB SNR, more than low 3 orders of magnitude of probability of 4 likelihood ratios of mistake, 2 mismarks of each DSQ symbol.
When 21, this probability has reached 4 orders of magnitude.Specify the approximate representation of the LSB bit bit error rate (BER) by below equation:
Figure 10 is the simplification for showing the decoding nDSQ symbols exported with LSB and MSB according to the embodiment of the present invention
Figure.This diagram is only an example, and the example should not excessively limit the scope of claim.Those skilled in the art can recognize
Know many changes, replace and change.As shown in Figure 10, once as described above, by MSB demappings, just with ratio DEC1
Decode it, and decoded LSB with ratio DEC0.
It is to be understood that according to the embodiment of the present invention, can be by being embodied in various ways nDSQ.For example, for SNR
=21dB, it is possible to use following parameter:
BCH (511,367, t=16), m=9, dim=4, BL=52ns and Baud=39.5GHz
BCH (511,314, t=14), m=9, dim=3, BL=39ns and Baud=40.1GHz
For SNR=20.5dB, it is possible to use following parameter:
BCH (511,349, t=18), m=9, dim=3, BL=38ns and Baud=40.8GHz
BCH (1023,783, t=24), m=10, dim=3, BL=77ns and Baud=40GHz
Generally, when there is decoding failure, only one BCH has detectable decoding failure (that is, more than t+1 and small
In the mistake of 2t).For example, using dim-2 the retained solution code bit of good BCH decoders and again to each DSQ
When only 2 LSB bits of symbol re-start Wagner decodings, position belongs to failure BCH and DSQ parity check bit.At this second
The decoded original bit bit error rates of Wagner reduce by 1/ (dim-1) and become the bit error rate of 2D-DSQ32:
This treatment can reduce BER at BCH outputs:
For example, the treatment is processed with many BCH postponing as cost, there is provided extra SNR nargin.
Although the above is the complete description of specific embodiment, can be using various modifications, replacing structure and wait
Jljl.Therefore, it is illustrated and described above to be not construed as limiting the scope of the present invention being defined by the following claims.