The application is to be that October 27, application number in 2004 are 200480039667.7 (international application no PCT/US2004/035893), are entitled as the dividing an application of application for a patent for invention of " unified MMSE equilibrium and the multi-user detection approach that are used for cdma system " applying date.
Embodiment
In following detailed description, with reference to show the accompanying drawing that can implement therein embodiment of the present invention by diagramatic way.These embodiments are described in enough detailed mode, to enable those skilled in the art to implement the present invention.Should be appreciated that although various embodiments of the present invention are different, and needn't be mutually exclusive.For example, connect specific feature, structure or characteristic that same embodiment is described together,, in the situation that do not deviate from the spirit and scope of the present invention, can realize in other embodiments.In addition, should be appreciated that in each disclosed embodiment,, in the situation that do not deviate from the spirit and scope of the present invention, can revise position and the arrangement of single parts.Therefore, the following detailed description should be as not restrictive, and only with appending claims, define scope of the present invention, with the whole equivalent scope of entitle claim, comes together to explain rightly scope of the present invention.In the accompanying drawings, same numeral represents same or similar functional in all several views.
The known RAKE receiver of generally using is to be received from the optimal solution of the CDMA signal of additive white Gaussian noise (AWGN) channel for demodulation.Yet, to disturb in (MAI) environment (for example meeting with in cellular CDMA-system) at multiple access, RAKE receiver is not optimum, and in some cases, RAKE receiver is not far optimal solution.In recent years, dropped into huge effort and developed advanced cdma receiver technology, to improve the performance of cdma network.Multiuser Detection (MUD) technology is an embodiment of advanced person's cdma receiver technology, wherein a plurality of users of jointly demodulation of receiver (namely expecting subscriber signal and other interference user signals).By using the knowledge that is associated about the user with the united demodulation, the MUD technology can improve demodulation performance significantly.The channel equalization technique that is applied to chip (chip) field represents another kind of advanced receiver technology, and it is merely able to be applied to orthogonal CDMA.Key idea in chip-rate (chip-rate) balancing technique is balanced communication channel, makes it near unitary transformation (unitary transformation), therefore recovers the orthogonality of destroyed signal transmission in channel.By such operation, crosstalk (cross-talk) between (same base station) different user is reduced, because the orthogonality of impaired signed codevector (signature code) is recovered at least in part in channel.In cellular system, the method only can be applicable to the downlink user of active base station, because usually only have these users' ability and desired user quadrature.
In at least one embodiment of the present invention, provide the unified MUD equalization approaches that is used for demodulation direct sequence CDMA (DS-CDMA) signal.Described unified approach can generate various cost-efficient receiver demodulation techniques, described demodulation techniques can (be compared with RAKE receiver from for example low-cost linear minimum mean-squared error (MMSE) balancing technique, it can provide the lowest performance gain), to the MMSE MUD (it can provide maximum performance gain) of relatively high complexity.Between, can select the compromise of various performance/complexity.Hereinafter, will use term " equilibrium " to represent that " signature independently " optimum mean square error (MSE) processes, and therefore can be extended the signal that comprises from the non-orthogonal base station of other and desired user.
Fig. 1 illustrates the block diagram of exemplary receiver structure 10 according to embodiment of the present invention.As shown, receiver structure 10 can comprise with lower at least one item: antenna 12, radio frequency (RF) are to baseband subsystems 14, joint equalization and MUD unit 16 and channel decoder 18.Antenna 12 can receive compound CDMA signal from wireless channel, and described compound CDMA signal comprises the overlapped signal that is associated with a plurality of users in system.Except desired user, composite signal can comprise the signal component from other users with the related same base of desired user and/or other users of being associated with other base stations.Can use the antenna 12 of what form, for example comprise and bipolar, paster, spiral, array and/or other to comprise the combination of above antenna.In at least one embodiment, use diversity antenna technology.RF is converted to the base band form of expression to the signal that baseband subsystems 14 will receive from the RF form of expression.RF can comprise assembly (component) such as low noise amplifier, one or more RF filter, one or more Frequency Transfer Unit (such as frequency mixer etc.) to baseband subsystems 14, and/or the signal that will receive is transformed into any other required assembly of base band.As will be described in detail, joint equalization and MUD equipment 16 are operating as and use same MUD/ equalization approaches to detect data in the signal of the compound reception that is associated with desired user.The data that channel decoder 18 detects based on predetermined channel code decoding, in order to be desired user recovery user data.Can use any type of channel code.In at least one embodiment, do not use chnnel coding.
Fig. 2 illustrates the block diagram of the embodiment of joint equalization and MUD equipment 30 according to embodiment of the present invention.Joint equalization and MUD equipment 30 can be for for example receiver structure 10 of Fig. 1 and/or the cdma receivers with other architectures.As shown, joint equalization and MUD equipment 30 comprise: sampler (sampler) 32, time-tracking unit 34, despreader (despreader) 36 and joint equalization device/MUD despreading sequence generator 38.Despreader 36 comprises multiplier 40 and accumulator 42.The baseband sampling that time-tracking unit 34 causes sampler 32 to receive with the chip-rate sampling, to generate chip-rate sampling y
_{k}Despreader 36 is by multiply by the chip-rate sampling signal that (namely in multiplier 40) comes despreading to receive by the despreading sequence that the joint equalization device/MUD despreading sequence generator 38 generates, then in symbol period, result is sued for peace by (namely in accumulator 42), with the output at accumulator, generate the expectation user symbol
Then, the desired user symbol can be passed to channel decoder and decodes.As will be described in detail, in at least one embodiment of the present invention, can generate the despreading sequence by joint equalization device/MUD despreading sequence generator 38, described despreading sequence is combined equalization and MUD processing by this way, namely when reaching the despreading performance of raising, also solve the problem of computation complexity.
In at least one is realized, can at first by the active user with in system, be divided into two groups and generate the despreading sequence.First group (namely organizing 1) is comprised of for known user for receiver " hypothesis " its signature sequence, and second group (group 2) is the unknown for receiver by " hypothesis " its signature sequence, and the first and second consistent users of statistics of its first and second grades of statistics and actual signal, consists of.The number of users of group in 1 can represent with K, and organize number of users in 2 and can use L (or L+1, wherein+1 is used for the additive white noise of representative system) to represent.As will be described in detail, when K and L variation, can realize different receiver architectures and performance.Generally with the MUD type, process to treat the user who organizes in 1.Such processing causes the higher performance take higher computation complexity as cost usually.On the other hand, generally with balanced type, process to treat the user who organizes in 2, such processing causes having the lower-performance of low computation complexity usually.In a word, along with the number of users that is assigned to group 2 increases, overall performance will descend, and overall computation complexity will descend.
Should be appreciated that at least one embodiment of the present invention, the decision that is placed to which group about the specific user depends on that not necessarily in fact whether known this user's signature in receiver.In fact, under many circumstances, receiver can be known all active users' signature.On the contrary, decision can be based on trading off between the performance in receiver and computation complexity.That is, can make decision, treat in one way the certain user, and treat in another way other users, to reach trading off between performance and complexity.For example, in a kind of approach, all users that are associated with same base as desired user are placed in group 1, and all users that are associated with other base stations are placed on and organize in 2 (that is, even known their signature of receiver).By this way,, to the application that the user's that is associated with serving BS (base station that namely with desired user, is associated) MUD processes, can reduce whole computation complexity by restriction.Replacedly, can use other to be used for the user is assigned to the technology of two groups.In at least one embodiment, for the technology that the user is assigned to two groups, can change in time (for example, it can be defined by the primary user of receiver and/or revise).
As described above, at least one was realized, the value by K and L in the variation receiver, can realize various receiver architecture.For example, if K=1 and L=0 can realize (have optimize weight) well-known RAKE receiver.Max ratio combined (maximal-ratio combining, MRC) RAKE receiver can be used as the special case near 0 time when signal to noise ratio (SNR) and realizes (in all other cases, known MRC is suboptimum).In another embodiment, if the number of users-1 in K=1 and L=serving BS, the available MMSE equalizer of realizing routine.In yet another embodiment,, if K=is all active users, can realize full-blown (full blown) MUD receiver.In addition, by select in a different manner K and L from top embodiment, can realize various cost-efficient associating MUD balanced reception machine technologies, described various associating MUD balanced reception machine technologies can optimally disturb (ISI) and multiple access to disturb (MAI) with given complexity level (under the MSE criterion) between process symbol.By this way, can reach the compromise of performance/complexity.
In the following discussion, the general technology that is used for determining joint equalization device/MUD despreading sequence has been described.In cdma receiver, the signal of the reception of baseband equivalence can followingly represent:
$y\left(t\right)=\underset{k=0}{\overset{K-1}{\mathrm{\Σ}}}{s}_{k}\left(t\right)\⊗{h}_{k}\left(t\right)+\underset{l=0}{\overset{L-1}{\mathrm{\Σ}}}{n}_{l}\left(t\right)\⊗{g}_{l}\left(t\right)$ (formula 1)
S wherein
_{k}(t) be the DS-CDMA signal, suppose that their signature (is supposed s without loss of generality, for known for receiver
_{0}(t) be all from start to finish desired signal), h
_{k}(t) be the impulse response (i.e. k impulse response that the DS-CDMA signal propagates through comprises the transmitter and receiver filter effect) of k bar channel,
Expression convolution operation symbol, n
_{l}(t) be white-noise process, and g
_{l}(t) be impulse response arbitrarily, provide below its feature is described in.Make g
_{0}(t) equal the impulse response of filter for receiver.Therefore, second of equation 1 and first just white-noise process to baseband signal role (for example, the thermal noise of receiver chain (receiver chain)).Other n
_{l}(t) item is used for representing the DS-CDMA signal, and the signature of supposing them is unknown for receiver, use g
_{l}(t) represent their channels (supposing that described channel is known for receiver) separately.Should reaffirm, suppose that signature can be relevant to the hypothesis signature for the difference between unknown CDMA user for known CDMA user, because it is directly involved in cost/trade-off of performance.Be to be further noted that traditional random spread spectrum hypothesis and chip shaping (chip-shaping) is embedded into g
_{l}(t) true directly hint n
_{l}(t) time (temporal) white character.According to just defined, the overall signal of result
Consistent with random spread spectrum DS-CDMA signal in its first and second grades of statistics (described statistics is that the MSE analysis is required).In addition, independently to cause unlike signal (be s to the statistics between the data flow of different user
_{k}(t) and n
_{l}(t)) statistics between is independent.Therefore can make following two hypothesis for the signal in formula 1, described hypothesis is enough to be used in following all derivations:
Suppose 1Lack correlation between-different user:
$E\{{s}_{{k}_{1}}\left(t\right)\·{s}_{{k}_{2}}{(t-\mathrm{\τ})}^{*}\}=0,{\∀k}_{1}\≠{k}_{2},\∀\mathrm{\τ};$
$E\{{n}_{{k}_{1}}\left(t\right)\·{n}_{{k}_{2}}{(t-\mathrm{\τ})}^{*}\}=0,{\∀k}_{1}\≠{k}_{2},\∀\mathrm{\τ};$ (formula 2)
$E\{{s}_{{k}_{1}}\left(t\right)\·{n}_{{k}_{2}}{(t-\mathrm{\τ})}^{*}\}=0,{\∀k}_{1}{,k}_{2},\mathrm{\τ}$
Suppose 2The unknown DS-CDMA letter of time white character-its signature is led for to(for) receiver:
$E\{{n}_{k}\left(t\right)\·{n}_{k}{(t-\mathrm{\τ})}^{*}\}=0,\∀k,\∀\mathrm{\τ}\≠0$ (formula 3)
Be purpose for convenience, it is incoherent that the symbol sebolic addressing of supposing different user is the time, is again incoherent between the user.Yet, should emphasize, this hypothesis is just to convenient, and its solution that other symbols distribute of can directly deriving.
The simple format of formula 1 allows to separate by the various parameters and the derivation MMSE that occur in definition suitably, studies different receiver architectures.In the following discussion, at first for the model inference of formula 1, going out general MMSE separates.Then, illustrate and how can derive different receiver architectures from general solution.Being used for n symbol (uses
Expression) MMSE output can followingly be expressed:
${\hat{a}}_{0}\left(n\right)=\underset{j={m}_{1}}{\overset{{m}_{2}}{\mathrm{\Σ}}}{{w}^{*}}_{j}\left(n\right)\·y(j\·{T}_{s})$ (formula 4)
T wherein
_{s}That sampling interval at receiver place, (the described sampling interval is chip period T normally
_{c}Integer/), and j ∈ m
_{1}, m
_{1}+ 1 ..., m
_{2}The suitable predefine observation window of center around n symbol.Typically, (m
_{2}-m
_{1}) T
_{s}Has T
_{c}Multiply by the identical exponent number of spreading factor (SF), although slightly large.Exact value depends on the delay expansion of channel, and can be the definable parameter of user of do as one likes energy/complexity compromise left and right.When the delay ratio symbol duration of channel is long, may expect to make observation window span (span) wider than symbol.Under such setting, the approach of advising also alleviates the intersymbol interference (ISI) that is produced by channel.Concerning being used for the low spreading factors that becomes more and more popular of high data rate applications, this may have special importance.Use vector symbol, formula 4 becomes:
${\hat{a}}_{0}\left(n\right)={{\stackrel{\‾}{W}}_{n}}^{+}\·{\stackrel{\‾}{Y}}_{n}$ (formula 5)
As is well known,
The MMSE solution provided by following formula:
${\stackrel{\‾}{W}}_{n}=E{\{{\stackrel{\‾}{Y}}_{n}\·{{\stackrel{\‾}{Y}}_{n}}^{+}\}}^{-1}\·E\{{\stackrel{\‾}{Y}}_{n}\·{a}_{0}{\left(n\right)}^{*}\}$ (formula 6)
A wherein
_{0}(n) expression desired user symbol.Use formula 1 and hypothesis 1, it can be depicted as:
$E\{{\stackrel{\‾}{Y}}_{n}\·{a}_{0}{\left(n\right)}^{*}\}{\stackrel{\‾}{P}}_{\mathrm{eq}\_0}\left(n\right)\·E\left\{\right|{a}_{0}\left(n\right){|}^{2}\}$ (formula 7)
Wherein
It is the equivalent frequency expansion sequence (referring to following formula 8 and formula 9) of considering the expectation symbol of channel effect.Make c
_{k}(i) k user's of expression chip sequence and definition:
${P}_{\mathrm{eq}\_k}(t,n)=\underset{i=n\·\mathrm{SF}+1}{\overset{(n+1)\·\mathrm{SF}}{\mathrm{\Σ}}}{c}_{k}\left(i\right)\·{h}_{k}(t-i\·{T}_{c})$ (formula 8)
Then, component
Be counted as and minute other symbol a
_{0}(n) T of " equivalence " frequency expansion sequence of corresponding desired user
_{s}The sampling at interval (that is, substitution k=0 in superincumbent formula 8, t=m
_{1}T
_{s}, (m
_{1}+ 1) T
_{s}..., m
_{2}T
_{s}).This can be expressed as on mathematics:
${\left\{{\stackrel{\‾}{P}}_{\mathrm{eq}\_0}\left(n\right)\right\}}_{j}=\underset{i=n\·\mathrm{SF}+1}{\overset{(n+1)\·\mathrm{SF}}{\mathrm{\Σ}}}{c}_{0}\left(i\right)\·{h}_{0}((j+{m}_{1}-1)\·{T}_{s}-i\·{T}_{c})$ J=1,2 ..., (m
_{2}-m
_{1}+ 1) (formula 9)
Also can be expressed as with matrix notation:
${\stackrel{\‾}{P}}_{\mathrm{eq}\_0}\left(n\right)={H}_{0}\·{\stackrel{\‾}{C}}_{0}\left(n\right)$ (formula 10)
H wherein
_{0}Be the Toeplitz matrix, its i is capable, the j column element is h
_{0}((i+m
_{1}-1) T
_{s}-jT
_{c}), and the chip vector
I element be i chip c
_{0}(i).Notice that formula 7 is following for s by using to formula 9
_{k}(t) traditional DS-CDMA model draws from formula 1:
(formula 11)
Wherein
The integer part (and notation index) of expression x, δ (t) represents Dirac delta function, and notices:
${\stackrel{\‾}{Y}}_{n}={P}_{\mathrm{eq}}\left(n\right)\·\stackrel{\‾}{a}\left(n\right)+\stackrel{\‾}{e}\left(n\right)$ (formula 12)
P wherein
_{eq}Be
Matrix, its k row are
(that is, by getting t=m
_{1}T
_{s}, (m
_{1}+ 1) T
_{s}.., m
_{2}T
_{s}The vector of calculating formula 8 gained), and
It is the symbol that is associated from different " users "
Individual vector.
Due to edge effect (for example when observation window is crossed over over single symbol period, or when channel delay spread is non-zero), vector
May need to comprise several continuous symbols and be used for each user (for example each user uses n-1, a n and n+1 symbol).Otherwise the model of formula 12 may only be set up (proximal edge effect (upto edge effect)) approx.When channel delay spread in-less-than symbol cycle (normally such situation) and observation window when not too large, above the insignificant edge effect of three common sufficient to guarantees of continuous symbol.In this case, extra symbol is regarded as extra " user ", therefore
(for example, when using n-1, during n and n+1 symbol,
).As a result,
Can comprise correspondingly with those sampling instants zero, wherein related symbol does not affect.Finally,
Be vector, its element is at t=m
_{1}T
_{s}, (m
_{1}+ 1) T
_{s}..., m
_{2}T
_{s}In time, be
, from formula 12, can illustrate
Can followingly calculate:
$E\{{\stackrel{\‾}{Y}}_{n}\·{{\stackrel{\‾}{Y}}_{n}}^{+}\}={P}_{\mathrm{eq}}\left(n\right)\·\mathrm{Diag}\left\{E\right\{{\left|{a}_{0}\left(n\right)\right|}^{2}\},...,E\{{\left|{a}_{\stackrel{~}{K}-1}\left(n\right)\right|\}}^{2}\left\}\right\}\·{P}_{\mathrm{eq}}{\left(n\right)}^{+}+E\{\stackrel{\‾}{e}\·{\stackrel{\‾}{e}}^{+}\}$
$=\underset{k=0}{\overset{\stackrel{~}{K}-1}{\mathrm{\Σ}}}E\left\{{\left|{a}_{k}\left(n\right)\right|}^{2}\right\}\·{\stackrel{\‾}{P}}_{\mathrm{eq}\_k}\left(n\right)\·{\stackrel{\‾}{P}}_{\mathrm{eq}\_k}{\left(n\right)}^{+}+\underset{l=0}{\overset{L-1}{\mathrm{\Σ}}}{N}_{l}\·{T}_{l}$
(formula 13)
N wherein
_{l}N
_{l}(t) power spectral density, and T
_{l}The Toeplitz matrix, its i, the j element is:
${\left\{{T}_{l}\right\}}_{i,j}\≡{\∫g}_{l}\left(\right[i-j]\·{T}_{s}+t)\·{g}_{l}{\left(t\right)}^{+}\mathrm{dt}=\frac{1}{2\·\mathrm{\π}}\∫{\left|{G}_{l}\left(w\right)\right|}^{2}\·{e}^{\mathrm{jw}(i-j){T}_{S}}\mathrm{dw}$ (formula 14)
(chip-spaced) sampling of the chip-spaced of known matched filter output (that is, with the filter for receiver of transmitter shaping filter coupling, making its output with the chip-rate sampling) forms the various lower a that arrange
_{0}(n) abundant statistical estimate.Therefore, when having carried out the sampling of chip-spaced, following formula can substitution formula 14 be derived to simplify:
T
_{s}=T
_{c}(formula 15)
Because the receiver of reality realizes usually using the sampling rate higher than chip-rate (usually for time-tracking and/or simplify anti-aliasing (anti-aliasing) analog filter and the purpose of the demand of analog to digital converter (A/D)), such substitution may not expected.Yet demodulation itself is carried out (for example, RAKE branch road (RAKE finger)) with chip-rate rather than higher speed usually.Be also noted that filter for receiver is usually bandwidth 1/2T simply
_{c}Unit gain (unit-gain) low pass filter (LPF) of (namely monolateral) or other Nyquists (Nyquist) filter (such as root-raised cosine etc.) can illustrate from formula 15 in either case:
${T}_{0}\≡\frac{1}{{T}_{c}}\·I$ (formula 16)
Formula 16 reflects such fact, and namely the chip-spaced correlation sequence of filter for receiver is obviously that Kronecker delta function (is recalled g in these cases
_{0}(t) be set to the impulse response of filter for receiver, and as anticipation, formula 16 hints that simply white noise variance item is N after receiver filtering
_{0}/ T
_{c}).
, although with balanced and/or MUD is not directly related, check that in environment of the present invention RAKE receiver remains useful., in order to do like this, at the receiver place, suppose minimum knowledge quantity.That is, suppose the only signature of known desired user (being K=1), and unknown any knowledge about the interference that is modeled as white noise (being L=1).Also hypothesis is being used chip-rate sampling (being formula 15).Use formula 6,7,8,13 and 16, as seen, solution is provided by following formula:
${\stackrel{\‾}{w}}_{n}=E{\{{\stackrel{\‾}{Y}}_{n}\·{{\stackrel{\‾}{Y}}_{n}}^{+}\}}^{-1}\·E\{{\stackrel{\‾}{Y}}_{n}\·{a}_{0}{\left(n\right)}^{*}\}$
$={[\underset{k\∈{\mathrm{\Ω}}_{0}}{\mathrm{\Σ}}{\stackrel{\‾}{P}}_{\mathrm{eq}\_k}\left(n\right)\·{{\stackrel{\‾}{P}}^{+}}_{\mathrm{eq}\_k}\left(n\right)+\frac{{N}_{0}/{T}_{c}}{E\left\{{\left|{a}_{0}\left(n\right)\right|}^{2}\right\}\·}\·I]}^{-1}\·{\stackrel{\‾}{P}}_{\mathrm{eq}\_0}\left(n\right)$
(formula 17)
Ω wherein
_{0}Be the representative with desired user (be that symbol is ..., a
_{0}(n-1), a
_{0}(n), a
_{0}(n+1) ... the user) set (the obvious 0 ∈ Ω of " user " index that symbol is corresponding
_{0})).
Next, suppose E{|a
_{0}(n) |
^{2}The N of }＜＜
_{0}/ T
_{c}(that is, with total noise power, compare, the user power of first despreading is very little, usually is assigned the situation of the fraction of total transmitting power as each user in cdma system), formula 17 can be approximated to be under these circumstances:
$\stackrel{\‾}{w}\left(n\right)\≈\frac{E\left\{{\left|{a}_{0}\left(n\right)\right|}^{2}\right\}}{{N}_{0}/{T}_{c}}\·{\stackrel{\‾}{P}}_{\mathrm{eq}\_0}\left(n\right)$ (formula 18)
It is just the matched filter that desired user is signed that formula 18 is removed gain term (gain term and demodulation performance are uncorrelated), and with known RAKE receiver (it is actually matched filter), conforms to.
What is interesting is, notice that in formula 17, the accurate MMSE solution to artificial situation has alleviated from MAI (namely by first in the expression formula that brackets), and therefore expect to provide and compare better performance with RAKE receiver.Yet this receiver is more complicated than RAKE receiver, and the performance boost of expection with respect to the cost that increases is unworthy.
Except above-described conventional RAKE receiver, the simplest receiver structure is the chip-rate equalizer., in order to draw this equalizer, suppose that receiver alleviates the MAI of serving BS by equilibrium, and every other user in system is treated as white noise.Therefore, following parameters can be used for formula 1: K=1 (that is, supposing the only signature of known desired user of receiver) and L＞1 (and under such setting, L equals the number of users in active base station).Also suppose:
g
_{l}(t)=h
_{0}(t), l=1 .., L (formula 19)
That is, all interference signals all experience the channel identical with desired signal (this is the situation of all downlink users of serving BS).It (is n that other unconnected users of all and serving BS are modeled as white Gauss (Gaussian) noise
_{0}(t) the outer MAI in all residential quarters of modeling (cell)).Also hypothesis is being used chip-rate sampling (being formula 15).
Based on above content, to separate by these parameter substitution formulas 6,7 and 8 being provided optimum MMSE, result is as follows:
${\stackrel{\‾}{w}}_{n}=E{\{{\stackrel{\‾}{Y}}_{n}\·{{\stackrel{\‾}{Y}}_{n}}^{+}\}}^{-1}\·E\{{\stackrel{\‾}{Y}}_{n}\·{a}_{0}{\left(n\right)}^{*}\}$
$={[\underset{k\∈{\mathrm{\Ω}}_{0}}{\mathrm{\Σ}}{\stackrel{\‾}{P}}_{\mathrm{eq}\_k}\left(n\right)\·{{\stackrel{\‾}{P}}^{+}}_{\mathrm{eq}\_k}\left(n\right)+\underset{l=1}{\overset{L}{\mathrm{\Σ}}}\frac{{N}_{l}}{E\left\{{\left|{a}_{0}\left(n\right)\right|}^{2}\right\}}\·{T}_{L}+\frac{{N}_{0}/{T}_{c}}{E\left\{{\left|{a}_{0}\left(n\right)\right|}^{2}\right\}}\·I]}^{-1}\·{\stackrel{\‾}{P}}_{\mathrm{eq}\_0}\left(n\right)$
(formula 20)
T wherein
_{L}Correlation Toeplitz matrix (by formula 19 substitution formulas 14 are obtained) for channel.First representative competition of the expression formula that is bracketed again, is from MAI (combating self-MAI).If suppose that to compare (first despreading) user power very little with (first despreading) gross power, formula 20 is reduced to:
${\stackrel{\‾}{w}}_{n}=E{\{{\stackrel{\‾}{Y}}_{n}\·{{\stackrel{\‾}{Y}}_{n}}^{+}\}}^{-1}\·E\{{\stackrel{\‾}{Y}}_{n}\·{a}_{0}{\left(n\right)}^{*}\}$
$\≈{[\underset{l=1}{\overset{L}{\mathrm{\Σ}}}\frac{{N}_{l}}{E\left\{{\left|{a}_{0}\left(n\right)\right|}^{2}\right\}}\·{T}_{L}+\frac{{N}_{0}/{T}_{c}}{E\{\left|{a}_{0}\left(n\right){|}^{2}\right\}\·}\·I]}^{-1}\·{\stackrel{\‾}{P}}_{\mathrm{eq}\_0}\left(n\right)$ (formula 21)
$={[\underset{l=1}{\overset{L}{\mathrm{\Σ}}}\frac{{N}_{l}}{E\left\{{\left|{a}_{0}\left(n\right)\right|}^{2}\right\}}\·{T}_{L}+\frac{{N}_{0}/{T}_{c}}{E\{\left|{a}_{0}\left(n\right){|}^{2}\right\}\·}\·I]}^{-1}\·{H}_{0}\·{\stackrel{\‾}{C}}_{0}\left(n\right)$
Called formula 10 while wherein transforming to the second row of formula 21.Attention is vector by the chip of desired user to unique dependence of notation index.Therefore, can realize this receiver by relevant (but is-symbol the is uncorrelated) conversion of first Application channel:
${\stackrel{\‾}{Z}}_{n}\≡{{H}_{0}}^{+}\·{[\underset{l=1}{\overset{L}{\mathrm{\Σ}}}\frac{{N}_{l}}{E\left\{{\left|{a}_{0}\left(n\right)\right|}^{2}\right\}}\·{T}_{L}+\frac{{N}_{0}/{T}_{c}}{E\{\left|{a}_{0}\left(n\right){|}^{2}\right\}\·}\·I]}^{-1}\·{\stackrel{\‾}{Y}}_{n}$ (formula 22)
Followed by be simple despreader:
${\hat{a}}_{0}\left(n\right)={\stackrel{\‾}{C}}_{0}{\left(n\right)}^{+}\·{\stackrel{\‾}{Z}}_{n}$ (formula 23)
Understand in depth for some that obtain the operation of this receiver, notice T
_{L}Channel autocorroelation function (referring to formula 19 and formula 14).Therefore, the proximal edge effect can use following formula (at least for the Exponential Stability channel) approximate:
${T}_{L}\≅\frac{1}{{T}_{c}}{H}_{0}\·{{H}_{0}}^{+}$ (formula 24)
For for for example generally the chip-spaced multipath channel models of use carry out verification expression 24, simply with the second row of channel frequency response (index and) substitution formula 14, and carry out integration (recalling receiver filtering is restricted to integration only in the Nyquist band).After the variation of some integration variables, result is:
${\left\{{T}_{l}\right\}}_{i,j}=1/{T}_{c}\underset{k}{\mathrm{\Σ}}{h}_{0}\left(\right[k+i-j\left]{T}_{c}\right)\·{h}_{0}{\left({\mathrm{kT}}_{c}\right)}^{+}$
This formula production 24 is ignored the approximate of edge effect.Using should be approximate, and in fact the channel correlating transforms (depends on the constant-gain factor by following formula
Provide:
${[{H}_{0}\·{{H}_{0}}^{+}+\frac{{N}_{0}}{\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{N}_{l}}\·I]}^{-1}\·{H}_{0}$ (formula 25)
If compare in residential quarter MAI with the white noise of MAI outside the modeling residential quarter very large (namely
), formula 25 is reduced to the contrary of channel.In this case, the equalizer of formula 25 recovers the orthogonality of the user code that is destroyed by channel fully.On the other hand, if
(white noise that namely represents other area interference is leading factor), formula 25 is reduced to matched filter, and receiver is reduced to RAKE receiver (it is the optimal receiver under this set).For any other setting, the equalizer in formula 25 (or the equalizer in formula 22) will produce (in the MSE meaning) best trading off between alleviating the same area interference that is used for the white noise component that disturbs outside the modeling residential quarter.
Should be noted that, the CDMA balancing technique can also realize with the sef-adapting filter theory.That is, can sequentially operate and to the sampling application self-adapting filter that receives, rather than in sampling
Vector on operate with batch mode.
In some cases, receiver is known the interference spectrum of other base stations.Under these circumstances, considering that this extraneous information simultaneously, can obtain the chip-rate equalizer, thereby cause augmented performance.Therefore, following parameters can be used for formula 1:K=1 (that is, suppose receiver only know the signature of desired user) and L=L
_{1}+ L
_{2}＞1 (L wherein
_{1}Equal the number of users in active base station).Known all interference noise experience from active base station channel identical with desired signal.This can followingly represent:
g
_{l}(t)=h
_{0}(t), l=1 .., L
_{1}(formula 26)
All (not from Serving cell) other users all by with l=L
_{1}+ 1 ..., L
_{1+}L
_{2}.l corresponding signal modeling.Also hypothesis is being used the chip-rate sampling.
Solution is in this case still provided by the general framework of formula 20-23.Understand in depth for some that obtain this solution, suppose the sight of two base stations of simplifying.Therefore, have except formula 26:
g
_{l}(t)=h
_{1}(t), l=L
_{1}+ 1 .., L
_{1}+ L
_{2}(formula 27)
(namely all interference signals from interference base station all experience identical channel).Use the approximate of this setting and formula 24, can obtain following equalizer:
${[{H}_{0}\·{{H}_{0}}^{+}+\frac{\underset{l={L}_{1}+1}{\overset{{L}_{2}}{\mathrm{\Σ}}}{N}_{l}}{\underset{l=1}{\overset{{L}_{1}}{\mathrm{\Σ}}}{N}_{l}}\·{H}_{1}\·{{H}_{1}}^{+}+\frac{{N}_{0}}{\underset{l=1}{\overset{{L}_{1}}{\mathrm{\Σ}}}{N}_{l}}\·I]}^{-1}\·{H}_{0}$ (formula 28)
If main interference source is same residential quarter MAI, as before, formula 28 is reduced to the contrary of channel, the orthogonality of this restoring signal signature.On the other hand, if other residential quarters MAI and thermal noise are leading factor, formula 28 is reduced to the matched filter in coloured noise, and this is the optimal solution of this sight.Under any other arranged, equalizer provided the optimal compromise between these two kinds of receivers.
For the performance of inspection formula 28 with respect to the conventional practice of formula 25 improves, consider the sight of two base stations, wherein expect the steady fading channel of base station signal experience, and interference base station signal experience AR (1) channel model
Suppose that the interference base station signal is more much better than than the expectation base station signal.Because the steadily decline of active base station experience is unit conversion (unity transformation) (be that it will cut little ice, and therefore be equivalent to RAKE receiver) so conventional white noise equalizer is reduced to (collapse).The SNR of this receiver output place is:
${\mathrm{SNR}}_{\mathrm{White}\_\mathrm{Noise}\_\mathrm{Equalizer}}=\frac{E\left\{{\left|S\right|}^{2}\right\}}{E\left\{{\left|I\right|}^{2}\right\}\·\frac{1}{1-{\mathrm{\α}}^{2}}}$
E{|S| wherein
^{2}Total transmitting power at expectation base station place, E{|I|
^{2}Total transmitting power at interference base station place, and
Be interference signal experience impulse response square tap (tap) and.Now, the coloured noise equalizer will be considered the spectral shape of interference signal, and the equalizer that will generate the albefaction noise (is recalled E{|S|
^{2}The E{|I| of }＜＜
^{2}, so equalizer will be only be processed roughly noise), and followed by be the matched filter of noise whitening device (whitener).Therefore, use this model, the frequency response of equalizer is:
(1-α·Z
^{-1})·(1-α·Z)＝-α·Z+(1+α
^{2})-α·Z
^{-1}
Wherein first (1-α Z
^{-1}) be that the channel that disturbs BS is got contrary noise whitening device; Second (1-α Z) is the desired signal matched filter of " newly " channel of experience now.SNR (ignores at hypothesis E{|S| in this case
^{2}The E{|I| of }＜＜
^{2}Situation under the very little same residential quarter MAI of impact) can followingly represent:
${\mathrm{SNR}}_{\mathrm{Colored}\_\mathrm{Noise}\_\mathrm{Equalizer}}=\frac{E\left\{{\left|S\right|}^{2}\right\}\·[{\mathrm{\α}}^{2}+{(1+{\mathrm{\α}}^{2})}^{2}+{\mathrm{\α}}^{2}]}{E\left\{{\left|I\right|}^{2}\right\}\·[1+{\mathrm{\α}}^{2}]}$
With respect to white noise equalizer (and with respect to the RAKE receiver that produces in this case with the identical performance of white noise equalizer), the SNR of coloured noise equalization approaches gains and can obtain by the ratio that calculates two SNR expression formulas:
${\mathrm{SNR}\_\mathrm{Gain}}_{\mathrm{Pr\; oposed}\_\mathrm{Colored}\_\mathrm{Equalizer}\_\mathrm{Approach}}=\frac{1+4\·{\mathrm{\α}}^{2}+{\mathrm{\α}}^{4}}{1-{\mathrm{\α}}^{4}}$
Can see, near 1 the time, the SNR gain is near infinite (namely when interference channel becomes more coloured (darker sky (deeper null)), SNR gain increase) as α., although discussing, can see that this conclusion is still set up under complicated situation more in relatively simple embodiment environment.
Better relatively will consider same residential quarter MAI.In this case, the SNR of white noise equalizer is constant, but coloured noise equalizer SNR is:
${\mathrm{SNR}}_{\mathrm{Colored}\_\mathrm{Noise}\_\mathrm{Equalizer}}=\frac{E\left\{{\left|S\right|}^{2}\right\}\·{(1+{\mathrm{\α}}^{2})}^{2}}{E\left\{{\left|I\right|}^{2}\right\}\·[1+{\mathrm{\α}}^{2}]+2\·E\left\{{\left|S\right|}^{2}\right\}\·{\mathrm{\α}}^{2}}$
(with respect to white noise equalizer and RAKE receiver) SNR gain is in this sight:
${\mathrm{SNR}\_\mathrm{Gain}}_{\mathrm{Pr\; oposed}\_\mathrm{Colored}\_\mathrm{Equalizer}\_\mathrm{Approach}}=\frac{{(1+{\mathrm{\α}}^{2})}^{2}}{1-{\mathrm{\α}}^{4}+2\frac{E\left\{{\left|S\right|}^{2}\right\}}{E\left\{{\left|I\right|}^{2}\right\}}\·{\mathrm{\α}}^{2}}$
Again, near 1 the time, SNR gains near infinite as α.(notice that this solution is aimed at E{|S|
^{2}The E{|I| of }＜＜
^{2}Situation).Can analyze similarly other situations.
In order to realize that linear MMSE MUD receiver embodies structure, can use following parameter: K=k in formula 1
_{1}+ k
_{2}..., k
_{B}(k wherein
_{i}The number of users of i base station) and L=1.Also suppose:
${h}_{l}\left(t\right)={\stackrel{~}{h}}_{1}\left(t\right),l=1,..,{k}_{1}$
${h}_{l}\left(t\right)={\stackrel{~}{h}}_{2}\left(t\right),l={k}_{1}+1,..,{k}_{1}+{k}_{2}$
. (formula 29)
${h}_{l}\left(t\right)={\stackrel{~}{h}}_{B}\left(t\right),l=\underset{i=1}{\overset{B-1}{\mathrm{\Σ}}}{k}_{i}+1,..,\underset{i=1}{\overset{B}{\mathrm{\Σ}}}{k}_{i}$
Wherein
Represent the channel between i base station and desired user.Under such setting, separate and separate consistent with the full-blown linear MMSE for the cellular downlink environment.
The analysis of complexity that above-described full-blown linear MMSE multi-user detector is carried out shows that its amount of calculation the best part is matrix E
Generation, rather than the inverting of matrix.In addition, this computation complexity is directly proportional to number of users in system.The following effectively cost-performance tradeoff of this hint, be that some users use MUD and to be that remaining users is used balanced.In at least one embodiment, the strongest K user in selective system, and with linear MMSEMUD, process efficiently this K user.For these users, take with the similar mode of top formula 29 as h
_{l}(t) assignment.Use the equilibrium of lower complexity to process efficiently a remaining L-1 user, due to a described remaining L-1 user a little less than, so they are so unimportant.For these weak users, take with the similar mode of top formula 29 as g
_{l}(t) assignment.
In order to realize such receiver architecture, the parameter in formula 1 can followingly be selected: K is user's definable parameter, determines cost-trade-off of performance, and L=total number of users amount-K+1.Also suppose:
${h}_{l}\left(t\right)={\stackrel{~}{h}}_{i}\left(t\right)$ Or
${g}_{l}\left(t\right)={\stackrel{~}{h}}_{i}\left(t\right)$ (formula 30)
Suppose that wherein l user belongs to i base station, and as before,
Represent the channel between i base station and desired user.In a word, this approach provides very flexible and attractive cost-trade-off of performance.Use enough large K value, performance is approached the performance of full-blown MMSE MUD.In addition, the less K value for the efficient mechanism that is accompanied by selection the strongest (disturbing most) user, can reach with much lower complexity similar performance.
In at least one embodiment of the present invention, used the approach (namely based on the receiver of one group of matched filter output, rather than based on the receiver of the sampling that receives) of symbol level.In order to consider the symbol level approach, certain K * (m is all multiply by on the both sides of formula 12
_{2}-m
_{1}+ 1) rectangular matrix A (becoming when in general A is), as follows:
${\stackrel{\‾}{X}}_{n}\≡A\·{\stackrel{\‾}{Y}}_{n}=A\·{P}_{\mathrm{eq}}\left(n\right)\·\stackrel{\‾}{a}\left(n\right)+A\·\stackrel{\‾}{e}\left(n\right)$ (formula 31)
The special A=P that selects
^{+} _{eq}Produce matched filter output group, but other selections are also possible.MMSE separates and directly from derivation before, draws now, and is as follows:
${\hat{a}}_{0}\left(n\right)=\underset{k=1}{\overset{K}{\mathrm{\Σ}}}{{b}^{*}}_{k}\left(n\right)\·{x}_{k}\left(n\right)={{\stackrel{\‾}{B}}_{n}}^{+}\·{\stackrel{\‾}{X}}_{n}$ (formula 32)
And the MMSE solution is:
${\stackrel{\‾}{B}}_{n}=E{\{A\·{\stackrel{\‾}{Y}}_{n}\·{{\stackrel{\‾}{Y}}_{n}}^{+}\·{A}^{+}\}}^{-1}\·E\{A\·{\stackrel{\‾}{Y}}_{n}\·{a}_{0}{\left(n\right)}^{*}\}$ (formula 33)
Now by formula 7 and formula 13 substitution formulas 33 are provided solution.In the situation that inverting of those matrixes is Calculation bottleneck, top conversion dimension is from m
_{2}-m
_{1}+ 1 dimension is reduced to K dimension that may be much smaller.Use large K value, performance can approach the performance of full-blown MMSE MUD.In addition, use and to be accompanied by the least K value of strong user's efficient mechanism of selection, above-mentioned receiver can reach with much lower complexity with MMSE MUD similar performance.
In some applications, can adopt multi-code transmission (be unique user and assign a plurality of yards signatures, normally in order to improve this user's throughput).In order to reduce computation complexity, may expect receiver sheet is shown a front end filter public for all yards (in the time of can being become), and the despreader of one group of routine (wherein each despreader be tuned to a concrete code) subsequently.This structure draws from formula 6,7 and 13 at once, discloses basic operation
Public for all yards, and afterwards need to followed by
(certainly, this operation is for the code special use).
Can be easily with above-described receiver family and RAKE receiver combination., in order to observe this respect, notice that the crosscorrelation item (occurring) of formula 7 is actually RAKE receiver in the general receiver structure of formula 6.Fig. 3 illustrates the block diagram of the exemplary receiver structure 50 that can be used for representing MUD receiver family according to embodiment of the present invention.As shown, receiver structure 50 comprises: antenna 52, RF are to baseband subsystems 54, frequency spectrum white function 56, RAKE receiver 58 and channel decoder 60.Antenna 52, RF work with the identical mode of describing before with basic to baseband subsystems 54 and channel decoder 60.Frequency spectrum white function 56 is processed the baseband sampling to baseband subsystems 54 outputs by RF according to first of formula 6 right sides.Consequential signal is passed to RAKE receiver 58, and RAKE receiver 58 is according to second this consequential signal of processing on formula 6 right sides.The output of RAKE receiver 58 is passed to channel decoder 18 and decodes.When existed system that upgrading has built based on RAKE receiver, above-described technology is useful.
In some cases, may expect the approach that will propose and disturb the elimination combination.For example, when the coding gain that can use channel code (for example convolution code or Turbo code) is carried out the interference elimination, may be this situation.As a kind of embodiment of possible realization, consider high-speed downlink packet access (HSDPA) channel of wideband CDMA (WCDMA), wherein can distribute to a user at name a person for a particular job all flow signed codevectors of base station of special time.Its transmission because this user can decode, but usually can not decode other users' transmission, be significant so attempt to carry out coded interference elimination and use above-mentioned equilibrium and/or MUD approach to treat every other MAi component on same residential quarter MAI., by the component that the signal correction that removes simply and will eliminate joins, can obtain optimal solution (supposing perfectly to disturb and eliminate) in this case from the signal that receives.For example, in above-mentioned formula 28, the MAI that eliminates the expectation residential quarter means and uses N
_{l}Substitute
This is corresponding to hypothesis and signal
Optimum chip-rate equalizer in the situation that the MAI that is associated is ideally eliminated.In the HSDPA application, these signals are only the results to the multi-code transmission of desired user.After adopting Turbo decoding (and using its coding gain), can estimate the information symbol of original transmitted, and can be used for eliminating based on the interference of routine recodification, remodulates and subtraction technology.
In one embodiment, this can operate iteratively.At first, calculate as described above weight, and all many yards of demodulation.Then, total symbol stream (comprising the symbol from all many coded signals) is passed to channel decoder.Then recode, interweave and the output of remodulates channel decoder, with the copy (replica) that generates many coded signals.Next, carry out demodulation iteration for the second time with (except just copy demodulated and that deduct) all copies of many yards from the signal that receives.Therefore, in this demodulation iteration for the second time, deduct other disturbing effects of many yards, caused the demodulation performance that improves.Again, the signal of the demodulation of described many yards is passed in channel decoder, the output of wherein said channel decoder is used for constructing the copy of many coded signals, and this process continues iteratively until meet certain stopping criterion.In each iteration, the soft information (being symbol reliabilities) of self-channel decoder output in the future is combined in copy generation and subtraction mechanism, thereby only partly deduct the symbol (vice versa) with low reliability, in order to improve the convergence of this approach.This technology is specially adapted to Turbo code, is exactly wherein iteration on the original matter of decoding mechanism, and relates to soft metric (symbol reliabilities) and calculate.Fig. 4 illustrates the block diagram of the exemplary receiver structure 70 that can be used for realizing interference cancellation techniques according to embodiment of the present invention.As shown, receiver structure 70 can comprise with lower at least one item: antenna 72, RF are to baseband subsystems 74, joint equalization and MUD unit 76, and channel decoder 78.Antenna 72, RF be to baseband subsystems 74, joint equalization and MUD unit 76, and channel decoder 78 can be to work with the mode like the component class that Fig. 1 describes of being combined before.In addition, provide feedback path 68 to feed back decoded information from channel decoder 78, with in joint equalization and MUD unit 76 or recode, interweave in other places in receiver structure 70, heavily demodulation and interference subtraction.As described above, this can carry out with iterative process.
Fig. 5 illustrates block diagram for the illustrative methods 80 of cdma receiver according to embodiment of the present invention.At first near the active user receiver is assigned to first group or second group (frame 82).Use here present communicating by letter and user that will be considered in term " movable with producing (active user) " expression system in the testing process for desired user.Active user can be associated with serving BS or another base station.Be assigned to the user of first group and can comprise and for example suppose that its signature sequence is known user concerning receiver, and be assigned to the user of second group and can comprise that its signature sequence of hypothesis is unknown user concerning receiver.Can assign any criterion in criterion to assign the user with various user.In addition, the appointment criterion of using can be that the user is definable.
Next the distribution based on active user between first and second groups generates associating MMSE equilibrium/MUD despreading sequence (frame 84).For example, this can carry out as previously described, wherein by the K with different and L value, realizes various receiver architectures.Use subsequently the CDMA signal (frame 86) that associating MMSE equilibrium/MUD despreading series processing receives.This can comprise and for example baseband sampling be multiply by the despreading sequence, and cumulative this result in accumulator then, with the despreading expected data.Use the associating MMSE processing type that equilibrium/MUD despreading sequence is carried out to depend on usually when initial how active user assigns.For example, when only comprising that desired user and second group do not comprise the user for first group, can be as process the CDMA signal in RAKE receiver.When comprising that all active users and second group do not comprise the user for first group, can be as process the CDMA signal in MMSE MUD.When only comprising that desired user and second group comprise all with other active users of the related same base of desired user for first group, can be as process the CDMA signal in the MMSE equalizer.When respectively comprising a plurality of user for first group and second group, can carry out the combination that MMSE is balanced and MMSE MUD processes.Replacedly, can use other structures.
In some embodiments, can use a plurality of reception antennas.Can above-mentioned derivation be expanded in direct mode the situation of a plurality of antennas.Specifically, the general solution of formula 6 is still set up, but should increase institute's directed quantity to comprise a plurality of aerial signals.For example, in the situation that two antennas,
Two antenna weight vectors
With
And the signal that receives should comprise two aerial signals
With
In above-mentioned detailed description, various features of the present invention are combined in single embodiment together, to simplify the disclosure.This open method should be interpreted as reflecting such intention, that is, invention required for protection need to be than the more feature of the feature of clearly stating in each claim.On the contrary, as appending claims reflected, inventive aspect was in the state that lacks than whole features of above disclosed single embodiment.
, although in conjunction with some embodiment, described the present invention, should be appreciated that and can take modifications and variations and without departing from the spirit and scope of the present invention, as those skilled in the art, be readily appreciated that.Such modifications and variations are considered to drop in the boundary and scope of the present invention and appended claims.