Embodiment
One, main theoretical basis of the reform of Chinese economic structure of the present invention
(1) principle of overlapping frequency division multiplexing
Although provided by the present invention also is a kind of frequency multiplexing technique, but stronger overlapping of ratio orthogonal frequency division multiplex OFDM can be arranged between its sub-carrier band, therefore be referred to as overlapping frequency division multiplexing or nonopiate frequency division multiplexing (Overlapped Frequency Division Multiplexing orNon-orthogonal Frequency Division Multiplexing).Actively a kind of new coding bound of the formation of utilization does not concern that spectrum overlapping is more but spectrum overlapping is not regarded as the phase mutual interference in the present invention, and encoding constraint length is longer, and coding gain is higher, and spectrum efficiency is also higher.Under same threshold wanted to interfering signal ratio condition, its available spectrum efficiency will be far above prior aries such as higher-dimension modulation; Otherwise when equal spectrum efficiency, its required threshold SJR is more much lower than technology such as higher-dimension modulation, and particularly its threshold SJR improves more in random time-varying channel.Be different from addition orthogonal frequency division multiplex OFDM, sub-band of the present invention can be the broadband signal frequency spectrum, allows to present the selectivity decline, so itself can have stronger anti-fading ability.Need to prove at last for frequency division multiplex techniques such as conventional frequency division multiplexing FDM, orthogonal frequency division multiplex OFDMs, the multiplexing spectrum efficiency of system that can't improve itself, and the present invention is by the multiplexing spectrum efficiency that increases substantially system.
The below is overlapping as example take three carrier spectrums, and the principle of overlapping frequency division multiplexing is described in detail.
As shown in Figure 4, there is the frequency spectrum of three signals to overlap among the figure.Because frequency spectrum overlapped, any sub-carrier signal is wherein carried out the severe jamming that demodulation all will be subject to other adjacent sub-carrier signals by conventional method, it is absolutely not therefore carrying out correct demodulation by conventional method.But, the present invention allows everybody scrutinize Fig. 4 again, supposes that the spectrum width of three be re-used signal A, B, C among the figure is B
0Conspicuous, subcarrier spacing, namely frequency shift amount is relatively
Conspicuous, that is to say that three signal spectrums overlap.For the sake of simplicity, suppose that the spectral shape of three be re-used signal A, B, C is in full accord, phase characteristic is zero entirely, and modulation system all adopts the binary antipodal modulation, and symbol lengths is T
sSecond, each subcarrier-modulated signal bandwidth is B
0Conspicuous, three signal Complete Synchronizations.Because their frequency spectrum overlaps, and all will be subject to the interference of adjacent other sub-carrier signals, by conventional method, correct demodulation is absolutely not.But the present invention does not process the frequency spectrum of each sub-carrier signal not isolatedly, but with the unified consideration of these three sub-carrier signals, at this moment situation is just fully different, because at certain T
sIn time period, the data that signal A, B, C three transmit are nothing but one of eight kinds of situations as shown in Figure 3.
And corresponding reception signal spectrum is respectively eight kinds of D, E, F, G, H, I, J, K among Fig. 3 when disregarding noise, and its data and spectrum waveform are one to one fully.Equally, for any other overlapping tuple, can check, can prove fully also that on mathematics its data and spectrum waveform also necessarily have one-to-one relationship simultaneously.In the ordinary course of things, if the frequency bandwidth of each sub-carrier signal that is re-used is B
0Conspicuous, B here
0Should take into account the whole video stretching factors in the system (such as system frequency drift, Doppler broadening etc.), when making overlapping frequency division multiplexing, frequency shift amount is that subcarrier spacing is that Δ B is conspicuous mutually, and satisfies
(K-1)ΔB<B
0≤KΔB;K=1,2,...
Namely have the frequency spectrum of K adjacent sub-carrier signal together overlapped, that each sub-carrier signal transmits all is M=2
QMetamessage, the i.e. equal load Q=log of its each symbol
2The information of M bit, if L this overlapped subcarrier arranged in the system, it finally is 2 that the data that then transmit may make up
QL=M
LKind, also have 2 one to one with it
QL=M
LPlant spectrum waveform.Therefore to belong to any spectrum waveform just passable as long as find out the data combination that transmits at receiving terminal.Although spectrum overlapping has destroyed the spectrum waveform of single data symbol itself, destroy the one-to-one relationship between single data symbol and its spectrum waveform, but do not destroyed the one-to-one relationship between data symbol sequence and its spectrum waveform.The most important theories basis of institute of the present invention foundation that Here it is.Certainly work as M
LWhen very large, the complexity that how to reduce system is exactly a very important practical problem.The present invention will provide the algorithm of an optimum to address the above problem, and its complexity only is decided by M
KRather than M
L
Present let us ordinary circumstance, as shown in Figure 5.The be re-used spectrum width of each sub-carrier signal of setting is B
0Conspicuous; Symbol rate is
Symbol/second; Modulation level is counted M=2
QBit rate is
Bps; Then the spectrum efficiency of single sub-carrier system is
Bps/hertz; The bandwidth of system becomes B after adopting L subcarrier
0+ (L-1) Δ B also is increased to simultaneously and always deliver a letter
Bps, so its spectrum efficiency is:
And
Can find out, sub-carrier number L>>K, when namely it is enough large, along with the increase of overlapping spectrum number K, the proportional raising of the spectrum efficiency of system, but the level number of system do not resemble and be index law the higher-dimension modulation technique and increase, and increases and just be the algebraically rule.For example, work as Q=1, when namely each subcarrier adopted binary modulation, the level number of K repetition spectrum overlapping system was K+1, only is linear growth with K, works as Q=2, and namely each subcarrier adopts M=2
2During=4 yuan of modulation, the level number of K repetition spectrum overlapping system homophase I channel is K+1, and the level number of quadrature Q channel also is K+1, and the overall level number of system is (K+1)
2, also just being square-law with K increases.Obviously when differentiable level number one timing of channel, adopt the spectrum efficiency of overlapping Frequency Division Multiplexing system will be higher than higher-dimension (being many level) transmission system.If for example it can adopt the 64QAM modulation under high-speed mobile condition to certain wireless communication system, its level number M=64, the level number of homophase I and quadrature Q channel is
And have same, the overlapping tuple K=7 of the overlapping frequency division systems of QPSK of I, Q channel level number, but its each symbol load 14 bits, and the every symbol of 64QAM system load 6 bits only, its spectrum efficiency only has overlapping frequency division systems
In general, if original system can be supported the M-QAM modulation, then under the same level number, adopt the overlapping tuple of the overlapping frequency division systems of QPSK modulation
Its spectrum efficiency exceeds than M-QAM modulating system
Doubly.
(2) mathematical theory of overlapping frequency division multiplexing basis
A. the emission of overlapping Frequency Division Multiplexing system signal and reception
Be without loss of generality, supposing that information source is that equiprobability is memoryless, is Ts second after its symbol duration channel transmission, and the message of transmitting is transmitted in frequency domain with parallel mode, be provided with altogether L subcarrier in the system, each subcarrier is all occupied B behind modulated filter and channel broadening
0Conspicuous bandwidth for the sake of simplicity, supposes that the modulation system of each subcarrier and filter complex envelope characteristic are in full accord, at basic B
0There is K sub-spectrum of carrier overlapped in the conspicuous bandwidth.
The complex data sequence that it is transmitted is:
n=0,1,2,....,
Wherein:
In the formula:
I
N, l, Q
N, lAt t ∈ [nT
s, (n+1) T
s], namely in n symbol time interval its 1st (1=0,1,2 ..., L-1) the data message level symbol that transmits of the homophase I of subcarrier and quadrature Q channel.
Its multiple transmitted signal is:
In the formula:
Be normalization complex modulated signal envelope, its complex frequency spectrum is
f
0T
s>>1 or be positive integer;
Δ B is relative frequency shift amount (subcarrier spacing), and it satisfies:
(K-1)ΔB<B
0≤KΔB;
E
0Be every symbol transmit signal energy.
And the occupied total bandwidth B of system is:
B=B
0+(L-1)ΔB,
L+K-1 rather than L Δ B are arranged in total bandwidth B.
For the sake of simplicity, in the present invention our intersymbol interference of will ignore first being caused by the channel time diffusion, namely or the supposition channel spread without the time, though perhaps free diffusion, the bit duration after the diffusion is Ts second, then multiple connection is collected mail and number is:
In the formula:
Be the complex envelope of white Gauss noise, its power spectral density is N0 watt/hertz;
φ
NlIt is the additional phase shift that channel is introduced;
E
sThe receiving symbol energy, E
s=α E
0, α is fading channel;
Because channel fading may be frequency selectivity, might be inconsistent to different sub carrier so receive the complex envelope of signal, therefore in (2) formula to the complex envelope of different sub carrier with
Expression, and
If channel is flatness decline, then each
All identical.
Because the symbol-interference in the system in the not free territory, symbol-interference only appears in the frequency domain, therefore there is no need to study whole reception burst, only studies t ∈ [nT
s, (n+1) T
s], n=0,1 .... be that reception signal in the single symbol duration is just enough, then multiple connection is collected mail and number is:
Wherein:
Its frequency spectrum is:
Wherein:
Be
(1=0,1 ..., frequency spectrum L-1), their bandwidth is B
0
(8) in the formula
It is the multiple time varying frequency response function of channel, because the rate of change of any actual channel is only relevant with the movement velocity of physical medium, such as wind speed, the speed of a motor vehicle etc., they (only have Ultra-Low Speed communication to be only an exception) in most cases, all be much slower than the actual rate of delivering a letter, therefore, they all are slow sex change concerning the overwhelming majority's channel comparatively speaking, can think that it is substantially constant in second at a symbol duration Ts, therefore
In the mark t that becomes when just there is no need to write expression, but for different n namely for the interval Ts's of distinct symbols
Can and should be different with the variation of channel.Particularly exist
In, namely l (l=0,1 ..., L+K-1) in the individual Δ B, the complex frequency spectrum that receives signal is:
(9) formula obviously is a plural convolutional encoding operation relation, as shown in Figure 6 and Figure 7.
Wherein:
Be the frequency spectrum of noise in this frequency range, it is a white spectrum usually;
l,k∈{0,12,...,L+K-1} (10)
U (f) is the frequency domain unit step function.
(9) and front different from the span of l in (10) formula are various, and it goes out greatly K-1 than sub-carrier number L, this be since in overall system bandwidth the number of Δ B than the large K-1 of going out of L, but should be noted that when l>L-1
Simultaneously
Remaining problem is used Maximum likelihood sequence detection MLSD algorithm provided by the present invention (will elaborate) exactly, finds the solution [the nT at t ∈
s, (n+1) T
s] namely n symbol makes the total bandwidth B of following formula in system in the duration in, namely
The data sequence u of interior minimum
n, n=0,1,2 ...
In the formula:
Expression all receives signal bandwidth;
|| ||
2The expression ● square mould.
(12) physical meaning of formula is at n symbol duration t ∈ [nT
s, (n+1) T
s] the most probable data sequence u of interior searching
n, make its corresponding spectrum waveform
With the spectrum waveform that receives signal
Near (Euclidean distance is minimum), namely find out and the immediate input data symbol of output spectrum the most.
Tap coefficient in B, the overlapping frequency division multiplexing frequency domain shift register channel model:
As everyone knows, in frequency-selective channel, the frequency response function of channel
Not only relevant with frequency, and present change at random along with the difference of observing time, it is in certain bandwidth B after frequency shift
0Interior shape generally is as shown in Figure 8 vicissitudinous.When frequency domain was made convolution algorithm, if disregard spectrum modulation signal properties influence after the filtering, each channel tap system should be consistent, because their representatives is same frequency.Spectral property after considering signal filtering (they be fully known) is when affecting, and each channel tap coefficient should be the product of channel spectral property after the response of this Δ B scope and respective signal filtering, as shown in Figure 7,
When being subcarrier spacing (frequency shift amount) much smaller than the coherence bandwidth of channel, then basically determined by spectral property after the known signal filtering, and channel in the response of this " frequency " only as a proportion weighted factor, this moment when doing the discretization digital processing, tap coefficient
Can tighten into some numerical value rather than waveform.Reality is realized bringing convenience.
(3) tree graph of overlapping frequency division multiplexing, trellis (Trellis) figure and state diagram represent
A. the tree graph of overlapping frequency division multiplexing represents
The tree graph of overlapping Frequency Division Multiplexing system represents it is a kind of mode of the very vivid overlapping Frequency Division Multiplexing system frequency domain of expression Input output Relationship.Fig. 9 be exactly the Q=1 of a K=3 be the Input output Relationship figure of the overlapping Frequency Division Multiplexing system frequency domain of binary, the branch that makes progress among the figure represents input bit u
n=1, downward branch then represents input bit u
n=-1, corresponding coding output then is illustrated in the top of each branch.Thick line path representation list entries is u=[1 among the figure ,-1 ,-1,1 ... ]
T, corresponding plural convolutional encoding output waveform then is
....Scrutinize this figure, can find that be one to one fully between frequency domain input and output sequence.Do not have absolutely certain list entries corresponding with two or more output sequences, on the contrary is also true.Therefore spectrum overlapping does not destroy the one-to-one relationship between the input and output sequence in the frequency domain.If so detect just the error probability that can not appearance can not subtract again by sequence in frequency domain, certainly traditional just must have been abandoned by quilt by the symbol detection mode.If the length of sequence is fixedly the time, for example its length is L, and then to Q unit information source, possible sequence sum will be 2
QL=M
L, problem will be summed up as 2
QL=M
LThe test problems of unit's signal.Because each sequence equiprobability of supposition occurs usually in communication, therefore should adopt the Maximum Likelihood Detection criterion, when each sequence unequal probability occurs, should adopt maximum posteriori criterion.So, seem the Optimum signal detection problem of overlapping Frequency Division Multiplexing system to have solved, the way it goes theoretically, has any problem but implement.Because L is usually larger, so direct utilization maximum likelihood or maximum posteriori criterion will be very complicated.Because people are when the list entries equiprobability, decoding algorithm to convolution code, after deliberation decades, and overlapping Frequency Division Multiplexing system also can be counted as a plural convolution coder in frequency domain, many decoding algorithms of therefore relevant convolution code, for example say Fano and the various Stack algorithm in best (being the maximum likelihood function value) path of search in tree graph, basically can use for reference the detection that is applied to signal in the overlapping Frequency Division Multiplexing system through after putting into the melting pot.Because these algorithms are not real best maximum likelihood algorithm, can only be referred to as the maximum likelihood algorithm that is as the criterion, the present invention does not just plan to have introduced.Below we will introduce another kind of algorithm---Maximum likelihood sequence algorithm (MLSD).This is real maximum likelihood algorithm, needs to introduce first trellis (Trellis) figure and state (State) figure of overlapping Frequency Division Multiplexing system for this reason.
B. the trellis structure of overlapping Frequency Division Multiplexing system and state denotation of graph
Although tree graph can vividly describe the relation between input and the output very much, this figure at first is bad picture, and particularly along with the increase of L, it will be index expands, and is not easy to use very much, and therefore is necessary it simplification.Return Fig. 9, can find after examining, this tree graph just becomes repetition after the 3rd, because every branch that gives off from the node that is labeled as a has same output, this conclusion is to node b, and c, d are correct too.They are nothing more than being following several possibility, as shown in figure 10.As can be seen from the figure can only transfer to (through input+1) node a and (through input-1) node b from node a, b can only arrive (input+1) c and (input-1) d simultaneously, c can only arrive (input+1) a and (input-1) b, and d can only arrive (input+1) c and (input-1) d.The reason that produces this phenomenon is very simple, because only have the individual symbol of adjacent K (be 3 specific to this example) just can form the phase mutual interference.So when frequency domain K bit data was input to channel, the 1st bit data had the earliest shifted out a rightmost frequency shift unit.Therefore the output of channel only is decided by the input of a front K-1 frequency data except depending on the input of existing frequency data.In general, for M=2
Q, i.e. Q dimension binary data input, as long as front K-1 Q dimension binary data is identical, they are just identical, corresponding output is just identical.Therefore among Fig. 9 (Q=1) behind the 3rd branch road, the node of every a of being labeled as just may be incorporated in together, same b, each node of c and d also may be incorporated in together, has so just formed a folding tree graph---trellis structure, as shown in figure 11.Among the figure regulation be input as+1 branch road represents with solid line, be input as-1 branch road and be represented by dotted lines.This is that downward branch has been inputted for-1 because can not stipulate simply that again branch upwards is+1 input after folding.
If get rid of the repetitive structure of trellis structure on frequency axis, can obtain a figure-state diagram (State Diagram) of further simplifying.State in the state diagram is to determine according to each node in the trellis structure, and namely each state is to be determined by the heavy Q dimension of the front K-1 that remembers in frequency domain at channel binary data bit.Therefore be the overlapping Frequency Division Multiplexing system of K to memory (constraint) length, its stable state number is 2 when binary is inputted
K-1Individual, when the input of Q dimension binary, be 2
Q (K-1)=M
K-1Individual.In addition in addition initial, final, front transition state and rear transition state, initial and end-state is (0,0) in this example; Front transition state is (0 ,-1) and (0,1) totally two; Rear transition state is (1,0), and (1,0) also totally two.State transitions relation initial and transition state is very simple, as long as noting initial and end-state must be entirely zero dummy status, and in front transition state, the former data that are stored in the channel contain zero, just have in the new data Q dimension binary+or-data, so they can only be come by a state transitions, and can to other 2
Q=M state (front transition state or stable state) shifts; For rear transition state, opposite with front transition state, former be stored in the channel be Q dimension binary+or-data, just have zeroly in the new input, therefore, they all can be by 2
Q=M state (rear transition state or stable state) shifts and comes, but can only be to a state transitions.Please note: in this example, Q=1, K=3, we are writing state a (1,1), and b (1,-1), c (1,1), d (1,-1) frequency relation of each information bit is arranged routinely from left to right the time, and for example bit 1 enters channel the earliest among the state b (1 ,-1) in frequency domain, but the bit that enters the earliest channel in the Q=1 channel model of Figure 12 but exists in frequency shift unit of rightmost, and its frequency relation is from right to left.This point please the reader must not be obscured.
We can find out that overlapping Frequency Division Multiplexing system is the finite state machine in the frequency domain, and its oriented state diagram can be described in the input/output relation of channel in the frequency domain fully.Because front K-1 the Q dimension binary information bit that each state representation is stored by channel, i.e. (K-1) Q information bit, the transfer branch road between state then represents the information bit that existing frequency is inputted.K=3 for example; The data bit of the binary channel input of Q=1 is ... ,-1,1,1 ..., then be to a state transitions in state diagram from the c state, because of c=(1,1), input again one 1 after, originally existed-1 in frequency shift unit of rightmost to shift out channel, and newly 1 of input entering channel, state transitions is to a=(1,1), the output of channel then is
, it is indicated on the transfer branch road from c to a.
The state diagram of overlapping Frequency Division Multiplexing system, as shown in figure 13, in general memory (constraint) length is that the Q of K ties up the binary input channel and has 2
Q (K-1)=M
K-1Kind of stable state, each stable state can to other 2
QIndividual state transitions, also can from other 2
QIndividual state transitions.In Trellis figure above-mentioned conclusion then report for: memory span be that the Q dimension binary input channel of K has M
K-1Plant different states, under stable case, each node outwards sends 2
Q=M branch road has again M branch road to be incorporated in this node simultaneously.
Trellis structure is very useful when research Maximum likelihood sequence MLSD algorithm.
(4) the Maximum likelihood sequence detection principle of overlapping frequency division multiplexing
Maximum likelihood sequence decoding algorithm in the convolution code can come the signal of overlapping Frequency Division Multiplexing system is detected through transforming to transplant.Below we still specifically introduce the MLSD algorithm as an example of binary signal example.We know for length is the Q dimension binary list entries of L, and its possible output sequence (possible path in Trellis or the state diagram) number is 2
QL=M
L, because L is usually very large, directly uses Maximum Likelihood Detection and will become very complicated.The essence of MLSD algorithm is maximum likelihood algorithm, but its complexity only is exponential increase with the memory span K-1 of channel, rather than is exponential increase with L.We suppose that the noise of channel is white noise for this reason, and the input data sequence that has the maximum likelihood function value in white noise channel should be the corresponding list entries in path that minimum Eustachian distance is arranged with the reception signal in tree graph or Trellis figure, i.e. selection the best
So that satisfy
In the formula:
For all receiving signal bandwidth;
But in Trellis figure, because each path periodically merges, there is no need to calculate likelihood function or the Euclidean distance of whole path fully.Because when the path merges, those paths that had relatively large Euclidean distance before merging can be removed fully.For example among Figure 11
The time have two paths to overlap for the first time at node a place, they are respectively:
(corresponding list entries is 1,1,1)
And
(corresponding list entries is-1,1,1).
Calculate respectively the Euclidean distance between this two paths and reception signal, stay one apart from the less person, be referred to as surviving path (Survivor Path), another distance relatively large person then removed.Therefore node a is write down first the surviving path that arrives it, for example say
And and receive Euclidean distance between signal
Also have two paths to overlap for the first time to node b equally, they are respectively:
(corresponding list entries is 1,1 ,-1) reaches
(corresponding list entries is-1,1 ,-1).Select one the path of relative minimum range to be arranged with receiving between signal, and write down this surviving path, for example say
And and receive Euclidean distance between signal
Node c and d are also done same processing, the results are shown in Figure 14.Surviving path among the figure all is relative minimum distance path.So we have obtained arrival node a, b, c, the relative optimal path of d and corresponding Euclidean distance:
In any judgement of this stage also bad work.
The time, the path that the equally respectively Euclidean distance between the different paths of each node of calculating arrival and reception signal, and selection has relative minimum range.For example to node a,
The time arrive a the path in original Trellis figure, have four, namely 1,1,1,1; 1 ,-1,1,1;-1,1,1,1;-1 ,-1,1,1.But in the calculating of phase I, first three branch road in second and A Third Way footpath is eliminated, so we can only make one's options in first and the 4th liang of paths.Need to calculate respectively them for this reason and receive Euclidean distance between signal.Notice that we do not need to calculate the Euclidean distance in whole path now, because noise is white noise, we only need to calculate
The time node a arrive
The time a between branch road and receive Euclidean distance between signal, add
It is the Euclidean distance between article one path and reception signal.Equally, we only need to calculate
The time node c arrive
The time node a between branch road and the Euclidean distance that receives between signal add r
C1It is the Euclidean distance between the 4th paths and reception signal.Between this two paths, eliminate the relatively large person of distance, and write down path and Euclidean distance thereof apart from less---
And
Certainly
With
Just can from memory, dispose.We also carry out same processing to node b, c and d.So continue, the every one-phase l that is calculating, to l (l=0,1,2, ..., L-K+1) nodes in stage separately the state of representative (being l each node in the stage among the Trellis figure) we only keep one with the path that receives signal Euclidean distance less, and write down this path Euclidean distance and corresponding path.
Figure 14 is a schematic diagram of this testing process, at the five-stage that calculates, namely
The time, each surviving path (being relative optimal path) is respectively:
At this moment the initial part in best path is-1 ,-1,1 entirely relatively.Therefore can enter a judgement:
Because the initial part of all relative optimal paths all is them, they are optimal path namely naturally.
If each surviving path does not have common initial part, then calculate and to continue, until they have had till the common part.Therefore the judgement of MLSD algorithm is at random, it might not adjudicate output for a long time, and judgement output also not necessarily by symbol, may once only have a judgement output, a plurality of judgement outputs also might once be arranged, but maximum judgement time-delay is length L+K-1 of Trellis figure.This is because for L subcarrier arranged, the multicarrier system that an adjacent K subcarrier spectrum is overlapped, because its Trellis schemes the longest L+K-1 that only has, and its end-state is complete zero (0,0 ...., 0), final each paths must merge, so the time-delay of Maximum likelihood sequence detection MLSD algorithm is at most the L+K-1 step.
Because this characteristics of MLSD algorithm can produce following two problems:
I) because the MLSD algorithm is to have common initial part that judgement output is just arranged with each surviving path, namely judgement will be through the time-delay of one section randomness.When L → ∞, the judgement time-delay is one of problem for the size of the probability of ∞ so.
Ii) the MLSD algorithm requires each state that two memories are arranged, and one is used for the Euclidean distance (Euclidean Distance) that storage arrives the relative optimal path of this state, and one is used for the relative optimal path that storage arrives this state.The capacity of this memory should select much two of problems that is so.
For first problem, verified system for L → ∞, its judgement time-delay is zero for the probability of ∞.
For in the Second Problem first, i.e. path Euclidean distance memory, in general because the existence of noise, no matter any paths, always it is increasing apart from receiving signal distance, from this point see this memory as if capacity should for ∞.But because our interested just relative distance between them, therefore after each calculating we can to make its maximum (or minimum) distance be zero, namely the distance of each surviving path is all deducted this maximum (or minimum) distance.Like this, storage will be the relative value of distance, its capacity is exactly limited.As for second, i.e. surviving path memory is got 5K or 4K is just passable according to the general length of experimental result, basically can ignore because surviving path is longer than the probability of 5K.In case if these memories were filled with and adjudicated and can adjudicate by force when also not carrying out this moment, namely the initial bit that minimum range is arranged is exported as judgement.Sometimes also available majority logic judgement is namely exported the majority in each surviving path initial bit as judgement.The very simple but slightly inferior properties of the equipment that adopts the latter is in first kind of way, but because the probability of judgement is very little by force, the performance loss that causes thus also is very little.
Different from other any communication technologys from the above mentioned, should process problem in frequency domain to the input of overlapping Frequency Division Multiplexing system.This just requires the receiver of this system must carry out at first to received signal continuous fourier-transform (FT) or discrete fourier-transform (DFT) or fast Fourier transformation (FFT), and then all computings are all carried out in frequency domain.Process problem in frequency domain, often make many former problems of processing more complicated in time-domain become very simple, such as filtering operation etc.People generally believe that spectrum overlapping can produce serious phase mutual interference, in fact overlappingly not only can not produce interference, are a kind of utilizable coding bound relation on the contrary, and overlapping more serious, coding bound is longer, and coding gain is higher.Certain this coding is a kind of encoding relation of self-assembling formation, and forced coding restriction relation not necessarily.And suitable coding makes it to form optimum encoding relation in the multiplexing cooperation of spectrum overlapping, can further improve systematic function.
Theoretical proof and Computer Simulation have been verified: in random time-varying channel, for fixing frequency shift amount (subcarrier spacing) Δ B, the available sub-carrier signal bandwidth B of widening
0Way increase overlapping tuple K, and along with the increase systematic function of K is become better and better, transmission reliability is more and more higher.When K → ∞, random time-varying channel will be transformed into the best perseverance of channel performance and participate in property white gaussian channel, i.e. awgn channel.Therefore in random time-varying channel, as long as system linearity degree and transmitting power are secure, just can improve simultaneously with the way that increases basic modulation signal bandwidth and spectrum overlapping tuple K spectrum efficiency and the transmission reliability of system.Certainly the complexity of system's processing also can increase thereupon.
Two, specific embodiments of the invention
Embodiment 1
Present embodiment provides a kind of method of overlapping frequency division multiplexing, and concrete steps are as follows:
Step 1: according to given channel parameter, system parameters determines the most basic some design parameters:
Channel parameter: mainly contain the maximum time diffusing capacity Δ (second) of channel or the coherence bandwidth of channel
(conspicuous); The peak frequency diffusing capacity of channel
The coherence time of (conspicuous) or channel
(second);
System parameters: mainly contain system bandwidth B (conspicuous); Requirement to spectrum efficiency; The linearity etc.;
Design parameter: mainly contain
Basic modulation level is counted M=2
Q, wherein Q is the information bit of every modulation symbol institute load.Because under same spectrum efficiency, system complexity and M are irrelevant, can suitably select as the case may be;
The basic symbol length T
s(second), basic spectrum modulation signal width B
0(conspicuous);
In order to reduce system complexity, the present invention advises T
s>>Δ;
B
0Select larger, T
sSelect longlyer, in random time-varying channel, can automatically produce the implied diversity gain, improve systematic function.Wherein:
Hidden frequency diversity tuple
Hidden time diversity tuple
Wherein
The minimum positive integer of expression.
Relative frequency shift amount (subcarrier spacing) Δ B or spectrum overlapping tuple K:
Less Δ B, thus larger K can improve the spectrum efficiency of system, but system complexity and allow all corresponding increases of level number will decide according to actual conditions and needs, and its fundamental relation is as follows:
(K-1)ΔB<B
0≤KΔB
Wherein: B
0In except basic modulation signal bandwidth, also should comprise system's maximum frequency shift and maximum dobla (Doppler) frequency expansion amount equifrequent the expansion factor.
At the coherence bandwidth of Δ B much smaller than channel
The time, the shift tapping coefficient in the overlapping frequency division systems channel model
To be punctured into the sample value of frequency spectrum after the modulation signal filtering, and channel in the response of this frequency only as a weighted factor.Otherwise the shift tapping coefficient in the overlapping frequency division systems channel model
To be some spectrum waveforms.
System subcarrier sum L:
Because system bandwidth B=B
0+ (L-1) Δ B, then
When giving fixed system and channel parameter, design parameter B
0, Δ B, K, L are that mutually restriction is closely connected together, should repeatedly select and carry out optimal design according to actual conditions.
The spectrum efficiency η of system is:
Give regularly at K, thus the less less B of Δ B
0, will cause larger L, higher η has higher requirement but L crosses conference to the system linear degree;
Too small B
0Naturally the hidden frequency diversity tuple that can cause system
Descend.But as long as system bandwidth B is enough wide, when design, just there is no need to take into account it.Because we still can be by the hidden frequency diversity tuple that interweaves, the technological means such as coding improves system, because system is sayed, its implied diversity tuple is decided by
Rather than
Just the latter understands self-assembling formation, and the former need just can obtain through additional technological means.
When each subcarrier-modulated signal spectrum width and modulation signal level number average were inconsistent, it is comparatively complicated that situation will become.But can select parameter to design according to mentioned above principle and method fully.
Step 2: according to the given characteristic of channel, system parameters and design parameter design the emission system of overlapping frequency division multiplexing.
Also be a parallel multi-carrier synchronization data transmission system because the multi-transceiver technologies such as overlapping frequency multiplexing technique and OFDM are the same, difference is that its demodulation is complete different from detection mode.That yes is essentially identical to its its transmitter architecture of each subcarrier and conventional digital communication transmitter.
Figure 15 is exactly the schematic diagram of overlapping frequency division multiplexing spread spectrum (containing a straight expansion or CDMA) system transmitter of realizing in time-domain.For non-spread spectrum or cdma system, wherein the spread spectrum arithmetic section can remove.Its n (n=0,1,2 ...) in the symbol time interval, the multiple computing that transmitter is realized is:
t∈[nT
s,(n+1)T
s] (13)
In the formula: E
0Be every symbol emitted energy;
Be normalization complex modulated signal envelope, it satisfies:
n=0,1,2,...
Be l (l=0,1 ..., the L-1) complex data that transmits of individual subcarrier.
The computing in its respective frequencies territory is:
In the formula:
It is this modulation signal of complex radical
Frequency spectrum, its bandwidth is B
0Conspicuous, namely
Be the frequency spectrum of each sub-carrier signal, its centre frequency is respectively f
0+ l Δ B (l=0,1,2 ..., L-1), bandwidth is B
0, namely
On engineering, the structure of Figure 15 is difficult to guarantee that its frequency domain computing is (14) formula, this is because the structure of its L filter, because each centre frequency is different, be difficult to guarantee in engineering the consistency of their complex envelopes, and its consistency can be brought great convenience in Receiver Design.Can change into frequency domain for this reason and realize that Figure 16 is exactly the basic block diagram of this realization.
Basic point among the figure is to form l=0 road complex modulated signal filtered spectrum with digital form first
In-phase component A
c(f-f
0) sequence, A
c(f-f
0) sequence
Phase shift is exactly
Quadrature component A
s(f-f
0) sequence, they will obtain l=1 behind the shift register of frequency domain, and 2 ..., the complex modulated signal filtered spectrum on other each road of L-1.Very simple in the frequency filtering computing, shift operation (being the frequency translation computing) is also very simple, and this has just guaranteed the consistency of each road signal spectrum complex envelope.
Step 3: it is synchronous to form symbol time in receiver, under synchronous condition, to the interval t ∈ of each symbol time [nT
s, (n+1) T
s], n=0,1,2 ..., reception signal formation multiple connection collection of letters frequency spectrum sequence
Its substep is as follows:
According to sampling theorem, select suitable sampling frequency, carry out to received signal digitized processing, form the time-domain Serial No. that receives signal.Digitized processing can also can be carried out in base band at intermediate frequency, is decided in its sole discretion by the designer fully.
The time-domain Serial No. carries out fourier-transform FT or discrete fourier-transform DFT (containing discrete fast Fourier transformation FFT) to received signal, under symbol time synchronous condition, to each symbol time duration [nT
sThe T of≤t<(n+1)
s], n=0,1,2 ..., the formation time segmentation receives signal spectrum
T ∈ [nT
s, (n+1) T
s], n=0,1,2 ...
Right again on frequency axis
Carry out the frequency domain segmentation, every section width is Δ B, namely
Wherein l (l=0,1,2 ..., L+K-1) section frequency spectrum be:
U (f) is the unit step function in the frequency domain;
l=0,1,2,......,L+K-1。
This is owing to have the L+K section with Δ B segmentation in total bandwidth B.
Step 4: actual channel is measured, found out channel linearity transfer function in the distinct symbols time interval
Valuation
T ∈ [nT
s, (n+1) T
s], n=0,1,2 ....
Valuation to channel transfer function can be used any method, for example utilizes special " pilot signal " to measure, or utilizes the information exchange of having adjudicated to cross the mode of received signal computing is calculated its valuation, even can find the solution its valuation with blind estimating method.
Step 5: utilize
step 4 to find
With known modulation signal
Frequency spectrum
Form overlapping Frequency Division Multiplexing system tap coefficient in the channel model in frequency domain
Wherein:
Because multiple connection is collected mail number at frequency separation
In branch road coding frequency spectrum, the complex frequency spectrum that namely receives signal component during noiseless is:
l,k=0,1,2,....,L-K+1;
This is just in the channel model trellis structure, l node, the plural convolutional encoding output of different conditions s different branch m.
Wherein: E
sBe the receiving symbol energy;
Be t ∈ [nT
s, (n+1) T
s] in the multiple the transmission of data of l subcarrier;
Tap coefficient is:
l,k=0,1,2,....,L+K-1
It is the unit step function of frequency domain;
Be l (l=0,1,2 ..., L-1) the reception modulation signal of individual subcarrier spectrum.
Step 6: count M=2 according to the basic modulation level that system adopts
Q, spectrum overlapping tuple K lists whole states of system; State comprises initial and end-state, front transition state, rear transition state and stable state, totally five classes.So-called state S is the modulating data that is stored in the frequency domain shift register channel model
Corresponding Q dimension binary data (+or-) or 0 data, wherein:
Initial and respectively one of end-state, they all are
(they refer to that all Q dimension data are 0 state);
Stable state has Q
K-1Individual, as shown in figure 17, they are respectively:
(they refer to that all Q dimension data all are states of binary (+or-) data).
Front transition state and rear transition state all have M+M
2+ M
3+ ...+M
K-2Individual.
Transition state refers to that the individual Q dimension data in front some (but being less than K-2) is zero state before so-called.
Transition state refers to that the individual Q dimension data in back some (but being less than K-2) is zero state after so-called.
Initial condition is only with to 2
QIndividual front transition state shifts, if K=2, then directly to 2
QIndividual stable state shifts;
End-state can only be from the front 2
QIndividual rear transition state shifts comes, if K=2, then directly from 2
QIndividual stable state shifts and comes;
The forward transition state can only be from the front state (initial condition or front transition state) shift and come, but can be rearwards 2
QIndividual state (front transition state or stable state) shifts; When the forward transition state only is present in node l among the Trellis figure<K-1.
Backward transition state can be from the front 2
QIndividual state (front transition state or stable state) shifts and comes, but a state (rear transition state or final transition state) shifts rearwards.When backward transition state only is present in node l among the Trellis figure>L-1.
Because always have new Q dimension binary or 0 yuan of new data to enter channel model, 0 yuan of K-1 Q dimension in front or two senior statesman's data are left channel model simultaneously, and Q dimension binary data has 2 at every turn
QKind the combination, and Q tie up 0 metadata only have one may, so the aforesaid state transfer relationship is just arranged.
Transition state is the peculiar state of overlapping frequency division multiplexing, and this is the difference with general convolution code or the corresponding finite state machine of Trellis code.
Step 7: according to the state transitions relation, draw state diagram, Trellis figure or the tree graph of system.And calculate the coding that respectively shifts branch road by (15) formula in the
step 5 respectively by the state transitions relation and export
That is:
Wherein: l ∈ (0,1,2 ..., L-K+1) the l sub-carrier frequencies of expression input, still
Behind l>L-1,
S represents to shift the state that branch road arrives at the l node;
M represents to arrive the path of this state, for front transition state m=1; Other state m=2
Q=M;
Because Trellis figure is until just finish behind the L-K+1 node, so l might be greater than L-1 in (18) formula, but when arriving the L-K+1 node, Trellis figure must shrink in final all-zero state.
Step 8: each stable state S is prepared two memories, and one of them storage arrives the surviving path U of this state S
S, l=[u
S, 0, u
S, 1..., u
S, l], l=0,1 ..., L+K-1, wherein u
S, lBe Q dimension binary data; Another then stores this surviving path U
S, lCorresponding coding is exported before the l node and is received the signal spectrum sequence
Between path Euclidean distance d
S, l, (l=0,1,2 ..., L-K+1).
Can temporarily use the memory of arbitrary stable state for transition state.Because stable state has M
K-1Kind, therefore every kind of memory all needs M
K-1Individual, need altogether 2M
K-1Individual memory.
Step 9: implement Maximum likelihood sequence detection MLSD in Trellis figure, its substep is as follows:
The path Euclidean distance that makes start node state (l=0) is d
0,0=0;
To the l node (l=1 ...., all the state S in L-K+1) calculate its whole m (m=1 or 2
Q=M) bar from last state transitions so far state the branch road code signal with receive the signal spectrum sequence
Between branch road Euclidean distance d
S, m(l, l+1).
To arrive the branch road Euclidean distance d of this state to every state S
S, m(l, l+1) and they set out separately path Euclidean distance d of state S
S ', l-1Addition forms m new path Euclidean distance, and therefrom selects reckling, as the path Euclidean distance d of l node state S
S, l, upgrade the path Euclidean distance memory that deposits this state S in.
Node l (l=1,2 ..., L-K+1), each state S is found out the corresponding surviving path U of its path Euclidean distance
S, lRenewal deposits the surviving path memory of this state S in.
To l+1 node iteron step 2,3,4, until the l=L+K-2 node.This moment, surviving path was inevitable only remaining one, and then the corresponding data sequence of this surviving path is exactly that our needed conclusive judgement is exported.
When sub-carrier number L is larger, in order to use short surviving path memory, its length can be decided to be 4K~5K, can check at any time the surviving path memory of each state this moment in substep 4, in case finding has identical initial part in the path that they keep, then the initial part that this is identical is vacateed corresponding memory space simultaneously as judgement output.
Capacity for the path Euclidean distance memory that reduces each state S, avoid overflowing, can make Euclidean distance minimum (greatly) person in path after each step is finished is zero distance, and it is relative Euclidean distance that the Euclidean distance memory of other each state is only stored with its difference (plus or minus).
Step 10: when spectrum overlapping tuple K is excessive, although step 9 has optimum performance, be that it can be found out in frequency domain and receive signal the path of real minimum Eustachian distance is arranged, but for excessive K, utilize step 9 will cause sequential detector too complicated, other rapid serial decoding algorithm that can consider to use for reference this moment in the convolutional encoding reduces the complexity of sequential detector, can be adapted to overlapping Frequency Division Multiplexing system but they must put into the melting pot.Its essence remains in frequency domain works, but the path that relative minimum Eustachian distance is only arranged with the reception signal of selecting, but the reduction of any Sequence Detection complexity all will be take sacrificial system thresholding signal to noise ratio as cost.
Above embodiment has illustrated the embodiment of overlapping frequency division multiplexing as example take the subcarrier of identical frequency range, yet overlapping frequency division multiplexing does not have specific (special) requirements to the frequency spectrum of multiplexed signals, signal spectrum shape, bandwidth on the different sub carrier can be identical, also can be different, the subcarrier spectrum interval is fixed.For the realizability of receiving terminal sequential detector, should limit the maximum bandwidth of signal on each subcarrier, such as counting B
0max, then can use B
0maxReplace B
0To calculate spectrum efficiency, status number etc.
Embodiment 2
Present embodiment provides a kind of system of frequency division multiplexing, as shown in figure 18, comprises digital signal dispensing device and digital signal processing apparatus, and the digital signal dispensing device comprises complex modulated signal generator and signal projector; Digital signal processing apparatus comprises signal receiver and received signal detector.
Complex modulated signal generator in the digital signal dispensing device is for generation of through the complex modulated signal after overlapped a plurality of subcarrier-modulated; Signal projector is used for sending this complex modulated signal.
Wherein, the complex modulated signal generator comprises serial to parallel conversion unit, modulated carrier frequency spectrum generation unit, carrier spectrum shift unit, multiplication unit, adder unit and Fu Shi inverse transformation unit as shown in figure 19.Below these several unit in the complex modulated signal generator are elaborated:
The serial to parallel conversion unit is used for inputting the data symbol sequence that serial bit stream is transformed to the multidiameter delay transmission.
The carrier spectrum generation unit is for generation of first or the in-phase component I of last subcarrier and the filtered spectrum signal of quadrature component Q.
The carrier spectrum shift unit, first that is used for the carrier spectrum generation unit is produced or the in-phase component I of last subcarrier and successively frequency displacement subcarrier spectrum of the filtered spectrum signal interval delta B of quadrature component Q, draw the in-phase component I of next subcarrier and the filtered spectrum signal of quadrature component Q, and with the in-phase component I of next subcarrier and the filtered spectrum signal frequency shift Δ B of quadrature component Q, go down successively to obtain the in-phase component I of all subcarriers and the filtered spectrum signal of quadrature component Q.
Multiplication unit, be used for the filtered spectrum signal of the in-phase component I of all subcarriers that will obtain through the carrier frequency shift unit and quadrature component Q and the generation of serial to parallel conversion unit the multidiameter delay data modulated signal in-phase component I with quadrature component Q complex multiplication, obtain the spectrum signal of every road signal after carrier modulation.
Adder unit is used for every road spectrum signal addition that multiplication unit is obtained.
Fu Shi inverse transformation unit, the spectrum signal that is used for adder unit is obtained converts time-domain signal to.
The signal receiver of digital signal processing apparatus is used for receiving the complex modulated signal after the overlapped a plurality of subcarrier-modulated of process that described digital signal dispensing device sends; Received signal detector utilizes the one-to-one relationship between the data symbol sequence that receives signal spectrum and transmission, detects data symbol sequence.
Wherein, signal receiver comprises symbol synchronization element as shown in figure 20, is used for forming sign synchronization in time-domain to received signal; And digital signal processing unit, be used for the reception signal in each symbol time interval is taken a sample, quantized, make it become the receiving digital signals sequence.
Received signal detector comprises the fourier-transform unit as shown in figure 21, is used for converting the time-domain signal that described signal receiver receives to frequency domain signal; The frequency segmentation unit is used for this frequency domain signal forming the actual signal subsection frequency spectrum that receives with spectrum intervals Δ B segmentation; The convolutional encoding unit is used to form the one-to-one relationship between the data symbol sequence that receives signal spectrum and transmission; And the Data Detection unit, be used for the one-to-one relationship according to the formation of convolutional encoding unit, detect data symbol sequence.
This convolutional encoding unit further comprises four parts as shown in figure 22: channel measurement unit, tap coefficient unit, trellis structure form unit and coding output unit.Wherein, channel measurement unit is for the linear transmission function of measuring channel
The tap coefficient unit is used for the linear transmission function according to channel
And subcarrier spectrum, the tap coefficient of generation encoder, i.e. overlapping subcarrier segmentation frequency spectrum
Trellis structure forms the unit, is used to form the frequency domain trellis structure of described system; The coding output unit is used for according to tap system and frequency domain trellis structure, produces the coding output of each state transitions branch road, namely receives the signal subsection frequency spectrum.
This Data Detection unit further comprises as shown in figure 23: surviving path memory cell, path Euclidean distance memory cell, branch road Euclidean distance memory cell, Euclidean distance addition unit, Euclidean distance comparing unit and decision unit.Below these several unit are elaborated:
The surviving path memory cell, be used for storage arrive the l node (l=1 ...., the surviving path of all the state S in L-K+1).
Surviving path Euclidean distance memory cell, be used for storage and arrive l node (l=1, ...., L-K+1) surviving path of all the state S in and actual Euclidean distance or the weighted euclidean distance of signal subsection frequency spectrum in frequency domain that receive, wherein, surviving path Euclidean distance memory cell can only be stored relative distance, namely by making minimum (greatly) person of path Euclidean distance or weights Euclidean distance be zero distance, it is relative Euclidean distance that the path Euclidean distance of other each state or weights Euclidean distance memory are only stored with its difference.
Branch road Euclidean distance memory cell, be used for storage for l node (l=1, ...., L-K+1) all the state S in, all each bar is from the so far branch road coding output of state of last state transitions and branch road Euclidean distance or the branch road weighted euclidean distance between the actual reception signal subsection frequency spectrum.
The Euclidean distance addition unit, be used for l node (l=1, ...., L-K+1) in the value addition of surviving path Euclidean distance memory cell of each state S ' that sets out of the branch road Euclidean distance memory cell of each state S and this state S, obtain a plurality of additive values.
The Euclidean distance comparing unit in a plurality of values that relatively the Euclidean distance addition unit draws, takes out minimum value, and upgrades value corresponding to surviving path memory cell neutral condition S with it.
Decision unit checks the described surviving path memory cell of each state, in case find in the path that they keep identical initial part is arranged, then the initial part that this is identical is vacateed corresponding memory space simultaneously as judgement output.
Above embodiment only is used for explanation the present invention, but not is used for limiting the present invention.