Wireless communications systems and methods

Information

  • Patent Grant
  • 10003394
  • Patent Number
    10,003,394
  • Date Filed
    Monday, July 20, 2015
    9 years ago
  • Date Issued
    Tuesday, June 19, 2018
    6 years ago
Abstract
Methods and systems that enhance interference cancellation in communication channels are described. Specialized time domain training sequences and simple cross correlation are used to obtain a channel estimate for use in stacked carrier beamforming and OFDM based spatial beamforming. In certain embodiments, a time domain preamble sequence is provided as an alternative to the conventional frequency domain preamble sequence. The use of a time domain preamble can increase channel estimation performance, facilitating cancellation of co-channel interference. Embodiments include generating a training sequence preamble in the time domain, computing a frequency domain response of the channel using an estimate of its time domain impulse response, and using the frequency domain response of the channel to calculate channel estimation from a cross correlation of data received from the channel against the training sequence.
Description
BACKGROUND OF THE INVENTION
Field of the Invention

The present invention relates generally to wireless communications and more particularly to systems and methods for cell-based wireless communications.


Description of Related Art

Many wireless systems require techniques allowing a base station to broadcast control messages to one or more users in an unsolicited fashion. The messages must often be sent in the presence of co-channel interference from other base stations. The requirement is driven in part by the increased user density and lower reuse factors, enabled by interference cancellation techniques such as spatial filtering and adaptive beam-forming. These techniques are in turn enabled by modern standards such as IEEE 802.16 (WiMAX). Furthermore many of the newer standards have adopted Orthogonal Frequency-Division Multiplexing (OFDM) technology, which provide unique challenges and opportunities to equipment manufacturers.


Issues include co-channel interference (CCI) that limit performance, especially at the cell edge. For WiMAX, as the performance of the frame control header (FCH) and the media access protocol (MAP) deteriorate, so does system reliability. The only conventional approach to resolving this issue is provided in the WiMAX standard and involves increasing code repetition times and frequency reuse. However, this approach causes a large control channel overhead, and the efficiency of the control channel becomes very poor.


SUMMARY OF THE DESCRIPTION

Methods and systems that enhance interference cancellation in communication channels are described. Samples are obtained from stacked carriers in a received signal and a data vector is created from the samples. Stacked carriers are selected using a steering vector received during synchronization of the receiver. The steering vector is calculated to obtain cancellation of interference from another receiver and is calculated based on time domain channel estimation. Specialized time domain training sequences and simple cross correlation are used to obtain a channel estimate for use in stacked carrier beamforming and OFDM based spatial beamforming.


In certain embodiments, a time domain preamble sequence is provided as an alternative to the conventional frequency domain preamble sequence. The use of a time domain preamble can increase channel estimation performance, facilitating cancellation of co-channel interference. In at least certain embodiments, a method and system for estimating performance of a wireless communications channel is disclosed. These embodiments include generating a training sequence preamble in the time domain that is uncorrelated with at least one other training sequence, computing a frequency domain response of the channel using an estimate of its time domain impulse response, and using the frequency domain response of the channel to calculate channel estimation from a cross correlation of data received from the channel against the training sequence. Other embodiments include channel estimation by computing beamforming weights for a plurality of subcarriers in the channel using second order cross correlation statistics from the frequency domain response of the channel to reduce effects of frequency dispersion.





BRIEF DESCRIPTION OF THE DRAWINGS


FIG. 1 shows a cell structure exhibiting co-channel interference;



FIG. 2 is a graph illustrating maximum ration combining (MRC) performance in a two interference cell example;



FIG. 3 is a graph comparing co-channel interference cancellation (CCIC) performance with MRC performance for two CCI cells at 0 dB, 10 dB;



FIG. 4 is a graph comparing CCIC performance with MRC performance for two CCI cells at 0 dB, 0 dB;



FIG. 5 depicts one example of a transmission procedure according to certain aspects of the invention;



FIG. 6 illustrates a channel estimation operation according to certain aspects of the invention;



FIG. 7 depicts a generic co-channel interference cancela tion (CCIC) receiver that implements one example of a CCIC procedure according to certain aspects of the invention;



FIG. 8 illustrates operation of a stacked carrier system;



FIG. 9 shows an example of a process for OFDM symbol generation;



FIG. 10 shows an example of a process for processing received signals;



FIG. 11 depicts one method for computing weights in a stacked carrier despreader;



FIG. 12 is a block diagram showing a space time adaptive beam-former;



FIG. 13 shows a channel estimation process;



FIG. 14 shows an inverse data autoregressive (AR) whitener estimation process;



FIG. 15 illustrates preamble structure and timing;



FIG. 16 is a graph comparing MRC and CCIC for one CCI cell;



FIG. 17 is a graph comparing MRC and CCIC for two CCI cells; and



FIG. 18 is a graph comparing MRC and CCIC for different coding repetitions.



FIG. 19 shows one example of a process for enhancing interference cancellation with channel estimation according to one embodiment of the invention.





DETAILED DESCRIPTION

Embodiments of the present invention will now be described in detail with reference to the drawings, which are provided as illustrative examples so as to enable those skilled in the art to practice the invention. Notably, the figures and examples below are not meant to limit the scope of the present invention to a single embodiment, but other embodiments are possible by way of interchange of some or all of the described or illustrated elements. Wherever convenient, the same reference numbers will be used throughout the drawings to refer to same or like parts. Where certain elements of these embodiments can be partially or fully implemented using known components, only those portions of such known components that are necessary for an understanding of the present invention will be described, and detailed descriptions of other portions of such known components will be omitted so as not to obscure the invention. In the present specification, an embodiment showing a singular component should not be considered limiting; rather, the invention is intended to encompass other embodiments including a plurality of the same component, and vice-versa, unless explicitly stated otherwise herein. Moreover, applicants do not intend for any term in the specification or claims to be ascribed an uncommon or special meaning unless explicitly set forth as such. Further, the present invention encompasses present and future known equivalents to the components referred to herein by way of illustration.


Although many of the techniques described herein assume the use of OFDM, but the invention is not limited to OFDM. For the purposes of this description, it is assumed that a broadcast channel can or must be transmitted over several narrow band channels, wherein a narrow band antenna assumption readily applies. However, the techniques described are not limited to broadcast channels, but can readily be applied to traffic channels as well.


The narrow band antenna assumption for the received signal at the subscriber unit can be written as:









x
k



(
n
)


=





q
=
1

Q








a
qk




d
qk



(
n
)




+


i
k



(
n
)




,




where xk(n) is the Mxl received complex data vector at sample index n and channel number k, aqk is the Mxl steering vector for source emitter q, frequency channel k, dqk(n) is the transmitted information symbol for emitter q, channel number k and sample index n, ik(n) is the interference vector seen on channel k and M is the number of antennas available to the receiver.


The n index may itself be due to additional frequency subcarriers or due to channel reuses over time. It may be assumed that, in the former case, the frequency subcarriers are closely enough spaced so that the channel can be considered constant. Thus the carrier spacing as indexed by k should be greater than any frequency diversity introduced by n.


Certain embodiments of the invention yield improvements in both control and traffic channels. Certain embodiments address issues associated with co-channel interference (CCI) that limit performance, especially at the cell edge. For the purposes of this description the example of IEEE 802.16e (WiMAX) media access protocol (MAP) will be used. For WiMAX, as the performance of the frame control header (FCH) and the media access protocol (MAP) deteriorate, so does system reliability. The current WiMAX standard provides no options for resolving this issue other than increasing the code repetition times and frequency reuse which can cause a large control channel overhead and significant degradation in control channel efficiency.


To permit multiple accesses for the broadcast channel the transmitters may be constrained such that a type of repetition code may be employed over the channel index. This is convenient because for example the 802.16 standard specifies the use of a repetition code for the downlink map. However, the use of the code according to certain aspects of the invention permits greatly improved performance in the presence of multiple co-channel emitter base stations. The repetition code, concept, referred to herein as stacked carrier, permits the receiver model to take the form of:











x


(
n
)


=





q
=
1

Q




a
q




d
q



(
n
)




+

i


(
n
)




,




(

eq
.




1

)








a
q



(





a

q





1




g
1








a

q





2




g
2













a
qK



g
K





)









i


(
n
)




(





i
1



(
n
)








i
2



(
n
)













i
K



(
n
)





)






(

eq
.




2

)








where a repetition spreading code given by g=(g1, g2, . . . gK)T is employed and where it is assumed that the same information symbol is used for all k, dqk(n)=gkdq(n). Although the logical carrier indices are numbered consecutively from 1 to K in Equation 2, the actual set of spread frequencies to which they map can be completely arbitrary. However the n indices imply a reuse of the same channel for reception of additional information symbols.


The stacked carrier concept allows for the cancellation of co-channel interferers greatly enhancing the performance over more traditional schemes such as diversity combining or maximum ratio combining. Moreover stacked carrier interference cancellation can be employed with minimal or no changes to the current WIMAX standard. Interference cancellation can be achieved in much the same way that interference cancellation is obtained in linear beam-forming, by placing nulls in the signal subspaces formed by the interfering emitters spreading codes multiplied by the wireless channel.



FIG. 8 illustrates a stacked carrier concept according to certain aspects of the invention from a receiver perspective. A set of subcarriers 80 is gathered at the receiver. Typically, in stacked carrier based systems, received data vectors can be created by sampling a vector from widely separated frequencies. The vectors form the columns depicted at 82. These vectors can be treated in a manner similar to data received from multiple antennas. It should be noted that, although the subcarriers are depicted here with constant interleaving, subcarriers may be chosen from an arbitrary set of frequency bins.


At the receiver it can be assumed that knowledge of the channel information is in the form of the steering vectors aq, although this knowledge may be uncertain and is typically obtained from synchronization/training sequences transmitted separately. In this regard, channel estimation procedures that can be obtained either in the time or frequency domains, depending on the channel model employed, will be described.


Receiver Processing


An optimum receiver can be derived by conditioning on the assumption of known steering vectors aq and then using an assumption that the transmitted symbols and the interference are Gaussian white noise. The statistics of aq, may then be considered as well as the possibility of non-white but unknown interference and then the possibility of more severe constraints on dq(n), which leads to the deployment of multi-user detection (MUD).


The receiver can be written in the further compressed form:

x(n)=Ad(n)+i(n) ,
A≡(a1, a2, . . . aQ)  (eq. 3)
d(n)≡(d1(n), d2(n), . . . dQ(n))T  (eq. 4)


Thus, using Bromberg notation and assuming that the interference is complex Gaussian, i(n)˜CN1MK(0, Rii) and that the information symbols are complex, unit variance, Gaussian white noise, d(n)˜CN1Q(0, I). From this it should be readily apparent that x(n)˜CN1MK(0, AAH+Rii) so that,

px(x|A)=π−MK|AAH+Rii|−1etr(−xH(AAH+Rii)−1x).

In certain embodiments, the conditional probability can be computed as pd(d(n)|x(n), A), from which the optimal least squares estimator for d(n) can be found. Typically, this quantity can be computed from:













p


(

d
|
x

)


=





p


(

x
|
d

)




p


(
d
)




p


(
x
)









=





π

-
MK







R
ii




-
1




etr


(


-


R
ii

-
1




(

x
-
Ad

)






(

x
-
Ad

)

H


)




π

-
Q




etr


(

dd
H

)





π

-
MK








AA
H

+

R
ii





-
1




etr


(


-



x
H



(


AA
H

+

R
ii


)



-
1




x

)










=




π

-
Q








A
H



R
ii

-
1



A

+
I





exp


(


-


(

d
-

d
_


)

H




(



A
H



R
ii

-
1



A

+
I

)



(

d
-

d
_


)


)










(

eq
.




5

)








d
_





(



A
H



R
ii

-
1



A

+
I

)


-
1




A
H



R
ii

-
1



x


,













where, etr(X)≡exp(Tr(X)) . Therefore,

d(n)|x(nCN1Q(w0Hx(n), (AHRii−1A+I)−1)
w0≡Rii−1A(I+AHRii−1A)−1

and thus the optimal expected least squares estimator is given by,

E(d(n)|x(n))=w0Hx(n).

This can also be written as:

w0=Rxx−1A
Rxx≡Rii+AAH.  (eq. 6)


It should be apparent that the estimate is independent and identically distributed over the sample index n. Note that it is possible to incorporate uncertainty in the estimation of A by assuming a tractable noise model for the distribution of A and integrating A out of the estimator in Equation 5.


Co-Channel Interference Cancellation (CCIC)


Code repetition and maximum ratio combining code repetition approach is used in WiMAX to enhance decoding performance. Maximum ratio combining (MRC) is used at a receiver to combine these repeated codes and improve the packet error rate (PER) or bit error rate (BER). MRC is an optimal combining method when the interferences comprise additive white Gaussian noise (AWGN). However, at a cell edge, co-channel interference (CCIs) is generally very strong and cannot be characterized as AWGN. In this case, MRC performance becomes very poor and PER may fall below acceptable thresholds. FIG. 1 shows an example of cell deployment configuration which includes a target cell 11 surrounded by neighboring cells 12-17. At cell edge 18 (for example), the Signal to Interference Ratio (SIR) can fall below 0 dB or worse. For example, in a system where (i) neighboring cells 12-17 are interference cells, (ii) frequency reuse is 1/3/1 and (iii) loading is 100% , a worst case measurement at the center of influence of three cells at circle 19 (for example), the SIR may be approximately −8.9 dB. FIG. 2 shows MRC performance at cell edge for a simulation of two interference cells where SIR is set for 0 dB and 10 dB for the cells. It is apparent that the MRC cannot reach the required 1% PER at cell edge. Although frequency reuse 1/3/3 can improve the cell edge performance, it carries the disadvantage of poor MAP efficiency. For example, in the case of frequency reuse 1/3/3 with code repetition 4, the MAP efficiency is only about 8%. Thus, the MRC approach can exhibit poor performance at cell edge.


Co-Channel Interference Cancellation by using Code Repetition


Certain embodiments employ CCIC methods to improve cell edge performance without losing MAP efficiency. Assuming that the data block Sik is repeated N times as

Xik=[SikSi+MkSi+2Mk. . . Si+(N−1)Mk]T,  (eq. 7)

where Sik is a coded and modulated data block with the block size M , Sik=Si+jk. i and i+j are the logical subcarrier indexes of the data block Si+jk, j=0, M, . . . (N−1)M and k is the index of the cells. After PUSC permutation, the logical subcarriers can be mapped to physical subcarriers as:

XiPUSCk=fPUSC(Xik),  (eq. 8)

where fPUSC( ) means PUSC permutation function and iPUSC is the index of the physical subcarriers. The received signal can be expressed as:











Z

i
PUSC

l

=





k
=
1

K




Ψ

i
PUSC

lk



X

i
PUSC

k



+

N

i
PUSC

l



,




(

eq
.




9

)







where l is antenna index at receiver, ΨiPUSClk is frequency domain channel transfer function from the k-th cell to l-th antenna at the receiver, NiPUSCl is an AWGN.


Assuming now that all the neighboring cells use the same data block size and code repetition numbers, and ΨiPUSClk can be estimated from the preamble symbols and expressed as {circumflex over (Ψ)}iPUSClk, k=1, 2, . . . K. Then, both received signal ZiPUSCl and estimated channels ĤiPUSClk are de-permutated:

Zil=fDE−PUSC(ZiPUSCl)  (eq. 10)
{circumflex over (Ψ)}ilk=fDE−PUSC({circumflex over (Ψ)}iPUSClk),  (eq. 11)

where fDE−PUSC( ) is the de-permutation function. The estimated channels play the same role as the steering vectors in (eq. 3). Thus:

Zil=[YilYi+Ml. . . Yi+(N−1)Ml]T  (eq. 12)
{circumflex over (Ψ)}ilk=[ĤilkĤi+Mlk. . . Ĥi+(N−1)Mlk]T  (eq. 13)

and, analogous to (eq. 3),













[




Y
i
l






Y

i
+
M

l











Y

i
+


(

N
-
1

)


M


l




]

=


[





H
^

i

l





1






H
^

i

l





2









H
^

i

l





K








H
^


i
+
M


l





1






H
^


i
+
M


l





2









H
^


i
+
M


l





K
























H
^


i
+


(

N
-
1

)


M



l





1






H
^


i
+


(

N
-
1

)


M



l





2









H
^


i
+


(

N
-
1

)


M



l





K





]

l






[




S
i
1






S
i
2











S
i
K




]


+

N
i
l





(

eq
.




14

)








If M≥K, Sik can be estimated according to the different criteria; for example, if it is assumed that RNil is the correlation matrix of the additive noise which is assumed to be AWGN and known, an MMSE estimate of Sik can be obtained as:










[





S
^

i
1







S
^

i
2












S
^

i
K




]

=


W
K



[




Y
i
l






Y

i
+
M

l











Y

i
+


(

N
-
1

)


M


l




]






(

eq
.




15

)








and Equation 6 can be rewritten:

Wk=(RNililKTHilK)−1HilKT  (eq. 16)

where











H
^

i
lK

=

[





H
^

i

l





1






H
^

i

l





2









H
^

i

l





K








H
^


i
+
M


l





1






H
^


i
+
M


l





2









H
^


i
+
M


l





K
























H
^


i
+


(

N
-
1

)


M



l





1






H
^


i
+


(

N
-
1

)


M



l





2









H
^


i
+


(

N
-
1

)


M



l





K





]





(

eq
.




17

)







If it is assumed that there exist L antennas at the receiver, then multi antennas can be used with code repetitions to cancel the CCI if LM≥K is satisfied. Here the number of the cancelled co-channel interference cells is K−1.


For example, if it is assumed that there are 2 receiver antennas and the code repetition is 2, up to 3 interference cells can be cancelled as flows:













[




Y
i
1






Y

i
+
M

1






Y
i
2






Y

i
+
M

2




]

=

[





H
^

i
11





H
^

i
12





H
^

i
13





H
^

i
14







H
^


i
+
M

11





H
^


i
+
M

12





H
^


i
+
M

13





H
^


i
+
M

14







H
^

i
21





H
^

i
22





H
^

i
23





H
^

i
24







H
^


i
+
M

21





H
^


i
+
M

22





H
^


i
+
M

23





H
^


i
+
M

24




]






[




S
i
1






S
i
2






S
i
3






S
i
4




]


+

[




n
i
1






n

i
+
M

1






n
i
2






n

i
+
M

2




]





(

eq
.




18

)







[





S
^

i
1







S
^

i
2







S
^

i
3







S
^

i
4




]

=


W
3



[




Y
i
1






Y

i
+
M

1






Y
i
2






Y

i
+
M

2




]






(

eq
.




19

)







W
3

=



(


R

N
i


+



H
^

l

3
T





H
^

i
3



)


-
1





H
^

i

3
T







(

eq
.




20

)








A simulated comparison of CCIC and MRC performance is provided in FIGS. 3 and 4, where the performances are obtained based on the parameters listed in Table 1. FIG. 3 shows CCIC performance compared to MRC in the case of 2 CCI cells at 0 dB, 10 dB. FIG. 4 shows CCIC performance compared to MRC in the case of 2 CCI cells at 0 dB, 0 dB.









TABLE 1







Simulation Parameter Setup








Parameter
Values












Channel BW
10
MHz








Duplex
TDD









Frame duration
5
ms


Sampling rate
11.2
MHz








Cyclic prefix










Symbol duration
0.1028
ms








Number of Cells
2 strong interference cells


DL permutation
PUSC


Number of antennas at BS
1


Number of antennas at MA
Up to 2


Channel
ITU Ped-B (3 km/h)


Channel Estimation
Ideal


Noise Variance Matrix
known










As will be appreciated from the comparisons, CCIC provides superior performance than MRC at cell edges.


With reference to FIG. 5, a procedure for transmitting according to certain aspects of the invention is described. In certain embodiments information bits 50 are grouped in a plurality of blocks at step 51, wherein the different blocks may have different block sizes. These information bits can comprise un-coded bits (i.e. pre-encoder bits), coded bits (after encoder), modulated symbols or un-modulated symbols. Different ones of coded or modulated blocks may employ coding and modulation schemes that are different from other coded or modulated blocks.


In one example of information bits that are modulated symbols, and assuming that Spk is the information symbol packet and k is a transmitter index or cell index, then:

Spk=[s1ks2k. . . sNk]  (eq. 21)

where, sik is the information symbol and N is the packet size. Spk can be grouped as:

Spk=SGk=[Sg1kSg2k. . . SgMk]  (eq. 22)

Sgik can be referred to as a block and gM is the number of blocks. Each block includes at least one symbol and each symbol in Equation (7) can belong to only one of the blocks. The block size of each block is the same as the number of symbols in the block but block size can differ between blocks. Moreover, different modulation and coding schemes can be used for different blocks. In certain embodiments, grouping order is provided such that a small block number corresponds to a small symbol index. For example, the block Sg1k=[s1ks2k. . . sB1k] and Sg2k=[sB1+1ksB1+2k. . . sB2k], etc. where B1 and B2 are the sizes of block g1 and g2.


Certain embodiments employ symbol interleaving (step 53) wherein two or more types of interleaving may be used for interleaving the symbols in each block. Each block can have a different interleaving function and symbol interleaving in each block may be independent of symbol interleaving in other blocks. However, in at least some embodiments, interleaving of symbols may be performed across plural blocks. Certain embodiments provide block interleaving wherein a plurality of blocks is interleaved.


The output of an interleaver can expressed as:

SIk=[SI1kSI2k. . . SIMk]  (eq. 23)

where IM=gM (see signal 54).


In certain embodiments, interleaved blocks may be repeated at step 55 K times, where K is the code repetition number at step 55. The repeated signal 56 can be expressed as in Equation (24):

SRk=[SI1kSI1k. . . SI1kSI2kSI2k. . . SI2k. . . SIMk. . . SlMk]=[SR1kSR2k. . . SRMk],   (eq. 24)
and
SRik=[SIik, SIik. . . SIik]  (eq. 25)

The number of SIik is determined by the code repetition times for the block. The repetition times for each block can be different, although in these descriptions, a common repetition time is assumed for all blocks.


Certain embodiments provide a permutation process at step 57 that includes regrouping SRk, inserting additional symbols or blocks into SRk and physical channel mapping based on certain predefined rules to obtain signal 58. In certain embodiments, the permutation step 57 includes interleaving. In one example, regrouping SRk can be considered to be a form of interleaving. However, certain embodiments employ a more flexible form of interleaving than permutation. Interleaving can comprise scrambling all of the symbols in Equation (23), scrambling all of the blocks in Equation (23), scrambling part of the symbols in Equation (23) and scrambling part of the block in Equation (23). Different scrambling rules can be provided including random scrambling or following some rules.


At the receiver, the received signal at l-th antenna is expressed as:










Z

i
p

l

=





k
=
1

K




Ψ

i
P


l





k




S

i
P

k



+

N

i
P

l






(

eq
.




26

)








where l is antenna index at receiver, ΨiPl is frequency domain channel transfer function from the k-th cell to l-th antenna at the receiver, NiPl is an AWGN.


In Equation (25), it can be assumed that at least one transmitted signal is a desired signal at receiver and all others are undesired signals which are referred to herein as “Co-Channel Interferences (CCI).” In seeking to cancel the CCI, it is assumed that all CCI and desired signals have the same block size and the same repetition number meaning that the size of block Sgik in Equation (22) is the same for any of k, k=1, 2, . . . K.


Referring now to FIG. 6, certain embodiments perform channel estimation at step 62, receiving signal 60 and outputting signal 64. In order to estimate the channel ΨiPlk, each of the neighboring cells is typically required to transmit a unique training sequence or unique preamble. Further at the receiver, the unique training sequence or unique preamble can be used for estimating the channel {circumflex over (Ψ)}iPlk, k=1, 2, . . . K. The channel estimation function can estimate all the neighboring cells' channels which are included in the target cell and CCI cells. The training sequence or preamble sequence can be a time domain or a frequency domain sequence and channel estimation can be performed in the time domain or the frequency domain.


Certain embodiments include de-permutation and/or de-interleaving functions for both received signal ZiPl and the estimated channel {circumflex over (Ψ)}iPlk, k=1, 2, . . . K. Assuming that the de-permutation or de-interleaving function to be fDEPk( ). Furthermore, different de-permutation and de-interleaving functions may be required since different cells can have different permutation and interleaving functions. Thus:

Zil=fDEP1(ZiCl)  (eq. 27)
{circumflex over (Ψ)}ilk=fDEPk({circumflex over (Ψ)}iPlk)  (eq. 28)

Assuming here that the target cell index k=1 and considering the correspondence with SRk in Equation (24) and (25):

ZRl=[ZR1lZR2l. . . ZRMl]T  (eq. 29)
ZRil=[YIi1lYIi2l. . . YIi(N−1)l]T  (eq. 30)
{circumflex over (Ψ)}ilk=[{circumflex over (Ψ)}R1lk{circumflex over (Ψ)}R2lk. . . {circumflex over (Ψ)}RMlk]T  (eq. 31)
and
{circumflex over (Ψ)}Rilk=[ĤIi1lkĤIi2lk. . . ĤIi(N−1)k]T  (eq. 32)

Regarding the subscript Iij, i refers to the block index used in Equation (23) and j refers to the repeated block index and j=0, 1, . . . , (N−1).


Certain embodiments provide a co-channel interference cancellation function (CCIC) that comprises channel matrix construction, weights calculation and equalization. The channel matrix can be obtained by using Equation (32):











H
^

Ii
lK

=


[





H
^


Ii





1


l





1






H
^


Ii





1


l





2









H
^


Ii





1


l





K








H
^


Ii





2


l





1






H
^


Ii





2


l





2









H
^


Ii





2


l





K
























H
^


Ii


(

N
-
1

)



l





1






H
^


Ii


(

N
-
1

)



l





2









H
^


Ii


(

N
-
1

)



l





K





]

.





(

eq
.




33

)








Thus:











[




Y

Ii





1


l











Y

Ii





2


l
















Y

Ii


(

N
-
1

)


l




]

=



[





H
^


Ii





1


l





1






H
^


Ii





1


l





2









H
^


Ii





1


l





K








H
^


Ii





2


l





1






H
^


Ii





2


l





2









H
^


Ii





2


l





K
























H
^


Ii


(

N
-
1

)



l





1














H
^


Ii


(

N
-
1

)



l





K





]



[





S
^

Ii
1







S
^

Ii
2












S
^

Ii
K




]


+

N
Ii
l



,




(

eq
.




34

)








where ŜIik is the estimated SIik under some criteria or the output of the equalization function. Assuming that

ŜIiK=[ŜIi1ŜIi2. . . ŜIiK]T,  (eq. 35)

the equalization can be expressed as Equation (36):

ŜIiK=WKlZRil  (eq. 36)

and the weight WKl can be obtained from:

WKl=custom characterEst(ĤIiK),  (eq. 37)

where custom characterEst( ) is a function consistent with certain of the criteria. The use of different criteria leads to different functions. For example, under MMSE criteria, Equation (37) can be rewritten as:

WKl=(RNIilIiKTĤIiK)−1ĤIiK.  (eq. 38)



FIG. 7 shows one example of a generic CCIC receiver. A processing chain for obtaining the estimated information symbols includes steps 70, 72, 74 and 76 while steps 71, 73, 75 and 77 describe the processing chain required to get side information such as the channel estimate and the construction of channel matrix due to the consideration of an arbitrary permutation map at the transmitter. The side information is ultimately used to obtain a beamformer/despreader which, after any interleaving due to error correction or logical grouping steps gives us our final information symbols. The logical grouping block extracts logical subchannels out of the raw data. While in general, channel estimation does not depend on subcarrier permutation, some information, such as the number of code repetitions may require an initial depermutation.


With reference also to FIG. 12, certain of the systems and methods described can be extended to include multi-antennas system when multi-antennas 120 are provided at a receiver (through stacked carrier filter bank 121). For example, equation (34) can be extended to a multi-antenna embodiments as:










[




Z
R
1






Z
R

2
















Z
R
L




]

=



[





H
^

Ii

1





K








H
^

Ii

2

K













H
^

Ii
LK




]



[





S
^

Ii
1







S
^

Ii
2












S
^

Ii
K




]


+

N
Ii
L






(

eq
.




39

)








Equations (35-38) can be easily extended to multi receiver antenna case. If LN≥K can be satisfied, (K−1) co-channel interferences can be cancelled.


Certain embodiments of the invention address the issues associated with CCI using a time domain preamble symbol instead of the current frequency domain preamble sequence for estimating the channel. By using a time domain preamble, channel estimation performance can be improved, and the obtained improvements can offset or cancel co-channel interference. MAP performance can also be improved significantly without increasing the overhead.


Multi-User Detection


In certain embodiments multi-user detection techniques may be deployed to improve performance of a receiver. Such techniques can provide useful result when a sufficient number of samples are available. Given the assumption that Q≤Q′, where Q′ is the total number of resolvable base stations, the diagonal components of AHRii−1A are estimates of the signal SINR and can help to determine which signals are the most promising. Each signal is linear beam-formed and passed through a demodulation/remodulation process. In one example, this latter process may involve recovering the nearest constellation point. The signal can then be removed from the received data, a new Rii can be estimated and the process repeated. The steps of the algorithm can be describes as follows:

    • 1. Compute w0≡Rii−1A(I+AHRii−1A)−1.
    • 2. Compute the best linear estimate {circumflex over (d)}(n)=w0Hx(n).
    • 3. Compute the information symbols “closest” to {circumflex over (d)}(n) by exploiting the constellation and or any error correcting codes, {tilde over (d)}(n) .
    • 4. Estimate the channel matrix for unmodeled users Â=<x(n){tilde over (d)}H(n)>Rdd−1.
    • 5. Remove the demodulated waveform from the received data for the strongest emitters, x(n)←x(n)−A{tilde over (d)}(n).
    • 6. Estimate a new interference covariance Rii←<x(n)xH(n)>
    • 7. Add a new pseudo-random steering vector(s) to the last column of the A matrix.
    • 8. Return to step 1.


In certain embodiments, additional steps may be taken to guarantee that newly acquired users are not converging to users that already exist in this blind acquisition process. One technique that can be used applies a constraint that keeps the matrix, AHRii−1A near to diagonal. Other embodiments employ one or more variations related to which set of emitters should remain captured within the A matrix and which emitters can be removed from the data x(n). Certain embodiments limit performance of these steps to a few iterations to avoid overmodeling the received data thereby causing divergence.


Although these methods and systems have been discussed in the context of processing downlink broadcast signals and downlink training signals, certain embodiments employ this same technology for uplink processing in various applications. FIG. 9 shows an example process for OFDM symbol generation with steps 90-96. Channel encoder (90) encodes the bits to be transmitted. Modulator (91) maps the encoded bits into the BPSK binary constellation. Mapping operation (92) maps the BPSK streams into real and imaginary parts of the complex plane. Code repeat operation (93) replicates a code block a fixed number of times depending on the measured SNR. Permutation operation (94) applies a base dependent permutation of the subcarriers. Insertion operation (95) inserts pilot training samples and guard band samples into the OFDM waveform in the frequency domain. The IFFT operation (96) performs an inverse FFT to bring the OFDM symbol back into the time domain. In one example, when time domain uplink training messages are transmitted from plural subscribers, a base station can learn the corresponding uplink channel (A matrices) for its in-cell users and possibly nearby out-of-cell users. As will be described in more detail below, this training process can also learn out-of-cell (or even in-cell) space-time interference covariance processes. The latter is particularly important for enabling spatial beam-forming, since it helps to model Rii.


Code Repetition in WiMAX


In accordance with WiMAX, the MAP information is repeated between 2 and 6 times after encoding. In certain embodiments, coding repetition can be exploited to cancel CCI. The repetition can be used as the stacking index k in Equation 35 and thereafter. For simple repetition, the transmit stacked carrier gains can simply be gk=1. In a single OFDM symbol, along with K repetitions, there is another factor of Mfec coding blocks each of Nfec subcarriers, making the total number of subcarriers utilized KMfecNfec. However, the ordering of the subcarriers can be randomized, scrambled according to different permutations specified in the WIMAX standard. Thereafter, pilots may be inserted to fill out the remainder of the OFDM symbol. FIG. 9 shows one procedure generating OFMD symbols.


In certain embodiments, the repetition code concept can be extended by replicating the forward error correcting (FEC) blocks with an additional scaling by the stacked carrier transmit gains gk. Depending upon the permutation employed, the replicated subcarriers can be treated as n indices in Equation 36 or as separate independent receive channels. The subcarriers that are most closely spaced can be considered as reuses of the same channel and therefore qualify as n indices. Widely-spaced subcarriers can be treated as independent channels. The stacked carrier concept can also be generalized to allow the subcarriers to be stacked to become dependent on the subcarrier index. This allows the frequency spacing between stacked carriers to vary slightly, without affecting the basic narrow band antenna model exploited in Equation 36. In an example based on the currently implemented WIMAX standard, a set of weights for each logical subcarrier in the replicated subchannel can be developed, since neighboring logical subcarriers are widely spaced after the permutation map. In this example, the spacing between subcarriers in the stacked carrier concept may also vary and must typically be tracked for other base stations as well as for the current base station in order to fill out the steering vectors with the correct channel estimate values. This approach can be especially useful when the permutation mapping for the downlink map is not allowed to vary from base station to base station.


However, in the example of the current WiMAX standard, a base dependent permutation of the subcarriers prevents perfect alignment of the K repetitions of a given information symbol from one base station to another. However, if the number of downlink MAP informational elements are bounded, this imperfection may be accommodated by noting that if a base station's subcarrier does not intersect with the set of K repetitions for a mobile station's assigned base station, then the channel response is effectively zero for that subcarrier and the channel response can be viewed as having a zero value for the spreading code for that subcarrier k, i.e. gk=0 in Equation 37. Therefore in the WIMAX example, for each set of K subcarriers to be despread using the linear weight vector, the permutation mapping from the neighboring interfering base station must be identified, and gk=0 can be applied to the non-intersecting subcarriers with gk=1 being applied to the intersecting subcarriers. The channel, itself in the form of steering vectors, can be estimated using time domain training sequences that employ fixed transmit weights and it can be assumed that any frequency spreading uses gk =1. Also in the specific case of dealing with the downlink MAP, it can be generally assumed that the unmodeled interference is white and proportional to the identity matrix, so that Riin2I in Equation 38.


If a repetition code of length K is used, standard array processing theory suggests that up to K−1 interferers can be cancelled and that the remaining degrees of freedom contribute to the array gain. Thus the array gain will be K-U, where K is number of degrees of freedom and U is the number of strong co-channel emitters. In one example of a receiver, the processing steps 101-104 are shown in FIG. 10. The receive data (100) is passed through an FFT (101). After the FFT, the pilots are stripped out in 102 and a depermutation occurs in 103, wherein the permutation structure is tracked and one or more sets of receiver weights is subsequently developed using Equation 36. To train those weights a clean copy of the estimated waveform is needed and thus the estimated modulated signal is inserted back into data matrix in 104.


Stacked Carrier Sparse Matrix Inversion Technique


With reference to FIG. 10, certain embodiments employ processes for handling systems conforming to unmodified versions of the WiMAX with based dependent permutations. The spreading factor K can be expanded to include either all of the available subcarriers or enough subcarriers so that the row dimension of the channel matrix A exceeds the column dimension in Equation 36. It will be appreciated that, in Equation 40 below, only a single reuse of the channel is considered and

x=Ad+i
{circumflex over (d)}=wHx  (eq. 40)
w=Rii−1A(I+AHRii−1A)−1.  (eq. 41)

The meaning of Q is altered in this section to consider all signals from all interfering base stations, so that dq represents the information symbol associated with the q′th informational element, and where 1:Q, includes all the informational elements from all base stations. Each informational element is spread over R pseudo-random frequency subcarriers, and the R frequencies chosen, are nearly random from one column to the next. It can be expected that R<<K. An interfering base station will have typically more than one informational element. Therefore, for the downlink MAP application, a single base station may be devoted to more than one column of the channel matrix A. Each qth column will typically be very sparse, having nonzero entries only at the R frequencies over which the transmitted information symbol dq is spread.


In certain embodiments, optimal linear despreading weights can be computed from Equation 41. Although both the row and column dimensions of A are potentially large, A is extremely sparse with only a fraction (R/K) of the entries being nonzero. In the limit as K approaches the size of the total number of subcarriers in the OFDM symbol, it can be expected that the column dimension is equal to C≡UK/R, where U is the number of interfering emitters. Thus provided that U<R, an over-determined system results that can be inverted for some signal processing gain. The interference covariance Rii will typically be modeled here as white noise and therefore a scalar of the identity, though a simple extension can also allow for a diagonal covariance matrix.


The inversion in Equation 41 can be achieved by exploiting the sparsity of A. Certain embodiments employ a QR-decomposition on A using Givens rotations such that A=QR, with Q a Unitary matrix and R an upper triangular matrix. Assuming for the purpose of this analysis that Rii2I,









w
=




A


(



σ
2


I

+


R
H


R


)



-
1








=




AR
+

-
1




R
+

-
H














R
+
H



R
+


=



σ
2


I

+


R
H


R







The R1 matrix can be found by employing a sparse QR algorithm (Givens rotations) to obtain the augmented matrix:







A
aug



(



R





σ





I




)









A
aug
H



A
aug


=


R
+
H




R
+

.







Since only the columns of w associated with the informational elements in the base station associated with a given mobile station are required, computations are further reduced to:

{circumflex over (d)}1=I1R+−1R+−HAHx,

where I1 is a C1×C truncated identity matrix with the top C−C1 rows removed corresponding to the out of cell informational elements and where it is assumed that the bottom C1 elements of d contain the in-cell informational elements. Rather than compute the inverse, back substitution can be used to implement multiplication by R+−1R+−H. Back substitution is preferable in this case since R+ should still be relatively sparse. The bottom of d is used so that back substitution can still proceed easily on the truncated {circumflex over (d)} output. At a high level, the steps 110-116 involved in the implementation of the despreader for this case are shown in FIG. 11. Although the basic block diagram is valid for any stacked carrier despreader, the example shown calls out components to be used when the channel matrix A is sparse.


Space Time Adaptive Beam-Forming


It is contemplated that certain embodiments may beamform directly in the frequency domain over a narrow frequency band or where the frequencies are spread but incorporated into the channel or steering vector. Wide band beam-forming in the time domain may also be used, where the beam-forming would be facilitated by the transmission of a known training sequence. For example, in OFDM systems training may comprise training of a beam-forming temporal process in the time domain which would permit conventional beam-forming over each subcarrier when transferred to the frequency domain.


The beam-forming problem in the time domain can be formulated as a least squares estimation problem which typically is equivalent to the maximum likelihood estimate under various assumptions. In one example, a time domain beam-former, or space-time adaptive beam-former (STAP) takes the form:









W
H



(
n
)


*

x


(
n
)






d
^



(
n
)










d
^



(
n
)




d


(
n
)












W
H



(
n
)


*

x


(
n
)








k
=

-

K
1




K
2






W
H



(
k
)




x


(

n
-
k

)





,





where d(n) is defined in Equation 37, and W(k) is a Ms×Q beam-forming matrix and where Ms=MK is the dimension of the received data vector x(n) . It can be assumed that, in the time domain, the waveform is optionally received over a set of band pass filters, one filter for each stacked carrier. Note also that the filter can be non-causal in order to accommodate the cyclic prefix typically used in OFDM. With the subscriber advancing its transmission gate, to remove propagation delays, important waveform information will be present in the cyclic prefix.


In certain embodiments, it is desirable to minimize the MSE objective function written as the time averaged error over n:

μ≡<∥{circumflex over (d)}(n)−d(n)∥2>n.

Setting the first derivatives to zero, the necessary conditions for a global optimum of the convex problem can be obtained. The derivatives are:












μ





W
*



(
k
)




=










k






x


(

n
-
k

)





x
H



(

n
-

k



)




W


(

k


)




-


x


(

n
-
k

)





d
H



(
n
)






n








=







k







R
xx



(

k
-

k



)




W


(

k


)




-


R
xd



(
k
)




,








where

Rxx(k′−k)≡<x(n−k)xH(n−k′)>n
Rxd(−k)≡<x(n−k)dH(n)>n.

For OFDM waveforms, the time shifts can be viewed as cyclic shifts. Setting the derivatives to 0, the necessary conditions force a solution of the form:

W=Rxx−1RXd,

where






W



[



W
H



(

K
1

)


,



W
H



(


K
1

-
1

)













,


W
H



(

-

K
2


)



]

H








R
Xd




[



R
xd
H



(

-

K
1


)


,



R
xd
H



(


-

K
1


+
1

)













,


R
xd
H



(

K
2

)



]

H








R
XX




(





R
xx



(
0
)






R
xx



(

-
1

)









R
xx



(


-

K
1


-

K
2


)








R
xx



(
1
)






R
xx



(
0
)









R
xx



(

1
-

K
1

-

K
2


)






















R
xx



(


K
1

+

K
2


)






R
xx



(


K
1

+

K
2

-
1

)









R
xx



(
0
)





)

.






Because of the Toeplitz structure of Rxx there exist fast algorithms for inverting the Rxx matrix and solving for W. Indeed the necessary conditions,










k











R
xx



(

k
+

k



)




W


(

k


)




=


R
xd



(
k
)







can be solved by taking the Fast Fourier Transform of both sides (after choosing K1+K2+1 to be a power of 2), so that in the Fourier domain with index m the necessary conditions become:

{tilde over (R)}xx(−m){tilde over (W)}(m)={tilde over (R)}xd(−m)  (eq. 58)

where {tilde over ( )} is used to indicate the Fourier transform.


A summary of the steps 121-127 required to obtain the linear STAP beam-former from the signal received at antennae 120 is provided in FIG. 12. The 122 and 123 blocks are correlations, which are defined and used in the equations listed above. The 124 block is the result of solving the linear equation 58. The 125 and 126 FFT blocks perform an FFT on the weights and the received data respectively. Note that in the example shown, one method of applying weights in the frequency domain is shown, which can be useful for OFDM processing. Since the weights represent a matrix convolution, applying the weights after performing an FFT separately on the received data and on the weights themselves will typically be more efficient.


Channel Estimation


Turning now to FIG. 13, the problem of channel estimation can be considered whereby the A matrix and the interference covariance used in the previous section can be obtained. The block diagram in FIG. 13, is a high level description of the channel estimation technique described in more detail below in the Channel Estimation Section. Block (130) is the raw receiver data, block 132 is the correlation operation between receiver data and the training sequence, block 136 constitutes the entire operation of estimate the channel and the whitening matrices, block 138 takes the FFT of the estimated channel, and/or the cross and auto correlations so that we can obtain the cross correlation statistic in block (139) as a function of frequency. In certain embodiments, channel estimation in the sampled time domain can be used to obtain a model of the vector impulse response over a set of discrete known delays. The Fast Fourier Transform (FFT) 138 of the estimated channel impulse response 136 and the interference whitening filter (e.g., 143 as shown in FIG. 14) can be taken to provide models for the channel and interference at any desired frequency subcarrier.


For the purpose of this description, it can be assumed that a known wideband training sequence 134 is transmitted from a given base station. This assumption may be generalized to include training sequences from several nearby base stations. Base stations that are further out can be modeled as quasi-stationary noise processes. For this scenario, there is a space-time model in the time domain:













x


(
n
)


=







j
=
1

J









h
j



(
n
)


*


s
j



(
n
)




+

i


(
n
)









=







j
=
1

J










k
=
0

K









h
j



(
k
)





s
j



(

n
-
k

)





+

i


(
n
)










(

eq
.




42

)








where h(n) is an M×1 impulse response sequence, sj(n) is a training sequence and i(n) is an interference plus noise process. Note that in this formulation hj(n) includes both the channel propagation and any transmit linear combining weights used at a given emitter.


The interference process may comprise a sum of many out of cell transmissions and can be modeled as a Gaussian MA process. Therefore, it is assumed that:







i


(
n
)


=




k
=
0

K








B


(
k
)




u


(

n
-
k

)












u


(
n
)


=


i


(
n
)


-




k
=
1

K








B


(
k
)




u


(

n
-
k

)






,





where it can be assumed that B(0)=I and that u(n) is a complex Gaussian noise process independent over n and u(n)˜CN1M(0, Ruu). It is further assumed that there are 1-1 mappings between the processes, x(n)⇄i(n)⇄u(n). From this the conditional distribution is:












p
u



(



x


(
n
)




u


(

n
-
1

)



,


u


(

n
-
2

)














u


(

n
-
K

)




)


=


π

-
M







R
uu




-
1




etr


(


-

R
uu

-
1





u


(
n
)





u
H



(
n
)



)
















u


(
n
)


=


x


(
n
)


-




k
=
1

K








B


(
k
)




u


(

n
-
k

)




-




k
=
0

K








H


(
k
)




s


(

n
-
k

)






,





(

eq
.




43

)








where

H(k)≡(h1(k), h2(k), . . . , hJ(k))
s(n)=(s1(n), s2(n), . . . , sJ(n))T.

The joint distribution can be written from this as:











p
x

(


x


(
n
)


,

x


(

n
-
1

)


,

x


(

n
-
2

)


,






)

=




p
u

(



x


(
n
)




u


(

n
-
1

)



,


u


(

n
-
2

)














)












p
u

(



x


(

n
-
1

)




u


(

n
-
2

)



,


u


(

n
-
3

)














)














=





q








π

-
M







R
uu




-
1



etr


(


-

R
uu

-
1





u


(

n
-
q

)

















u
H



(

n
-
q

)


)

,








where










u


(

n
-
q

)


=


x


(

n
-
q

)


-




k
=
1

K








B


(
k
)




u


(

n
-
k
-
q

)




-




k
=
0

K








H


(
k
)





s


(

n
-
k
-
q

)


.








(

eq
.




44

)







Therefore, for maximum likelihood (ML) estimation purposes the log likelihood is:










ρ
ML

=



q








(


-

log


(



R
uu



)



-

Tr


(


R
uu

-
1




u


(

n
-
q

)





u
H



(

n
-
q

)



)



)

.






(

eq
.




45

)








Although ML estimation is used in this example for channel estimates, it will be assumed that u(n−q)=0 for large enough q. For the environment of this example, this assumption is nearly valid, since the training sequences have finite extent. However, the primary reason for taking the assumption is to obtain a generalized version of the Yule-Walker equations, which are consistent and stable solutions to the channel estimation problem. ρML is differentiated by the unknown channel matrices to obtain the necessary conditions.


The necessary conditions for the estimation of H(k), and B(k) are:









q








x


(

n
-
q

)





s
H



(

n
-
k
-
q

)




=





m
=
0

K








H


(
m
)






q








s


(

n
-
m
-
q

)





s
H



(

n
-
k
-
q

)






+











m
=
1

K








B


(
m
)






q








u


(

n
-
m
-
q

)





s
H



(

n
-
k
-
q

)

















q








x


(

n
-
q

)





u
H



(

n
-
k
-
q

)




=





m
=
1

K








B


(
m
)






q








u


(

n
-
m
-
q

)





u
H



(

n
-
k
-
q

)






+











m
=
0

K








H


(
m
)






q








s


(

n
-
m
-
q

)






u
H



(

n
-
k
-
q

)


.












It will be appreciated that the inverse spatial covariance matrix Ruu−1 can be canceled from both sides of the necessary conditions. By changing variables n′=n−q and sum over n′ and divide both sides by the time support length of the training sequences, then:








R
xs



(
k
)


=





m
=
0

K








H


(
m
)





R
ss



(

k
-
m

)




+




m
=
1

K








B


(
m
)





R
su
H



(

k
-
m

)













R
xu



(
k
)


=





m
=
0

K








H


(
m
)





R
su



(

k
-
m

)




+




m
=
1

K








B


(
m
)





R
uu



(

k
-
m

)










where

Rxs(k)≡<x(n+k)sH(n)>n
Rss(k)≡<s(n+k)sH(n)>n
Rsu(k)≡<s(n+k)uH(n)>n
Rxu(k)≡<x(n+k)uH(n)>n
Ruu(k)≡<u(n+k)uH(n)>n

and where the time averages are taken over the support (non-zero elements) of s(n) and u(n) . Further, the convention that the data components x(n) are defined to be 0 when indexed outside of the collection interval may be adopted. This may be expressed as:

Rxv<CRvv,

where







R
vv



(





R
ss



(
0
)






R
ss



(
1
)









R
ss



(
K
)






R
su



(
1
)






R
su



(
2
)









R
su



(
K
)








R
ss



(

-
1

)






R
ss



(
0
)









R
ss



(

K
-
1

)






R
su



(
0
)






R
su



(
1
)









R
su



(

K
-
1

)


































R
ss



(

-
K

)






R
ss



(


-
K

+
1

)









R
ss



(
0
)






R
su



(

1
-
K

)






R
su



(

2
-
K

)









R
su



(
0
)








R
su
H



(

-
1

)






R
su
H



(
0
)









R
su
H



(

K
-
1

)






R
uu



(
0
)






R
uu



(
1
)









R
uu



(

K
-
1

)








R
su
H



(

-
2

)






R
su
H



(

-
1

)









R
su
H



(

K
-
2

)






R
uu



(

-
1

)






R
uu



(
0
)









R
uu



(

K
-
2

)


































R
su
H



(

-
K

)






R
su
H



(


-
K

+
1

)









R
su
H



(
0
)






R
uu



(

1
-
K

)






R
uu



(

2
-
K

)









R
uu



(
0
)





)







C


(


H


(
0
)


,

H


(
1
)


,








H


(
K
)



,

B


(
1
)


,

B


(
2
)


,








B


(
K
)




)








R
xv




(



R
xs



(
0
)


,


R
xs



(
1
)


,









R
xs



(
K
)



,


R
xu



(
1
)


,


R
xu



(
2
)


,





,


R
xu



(
K
)



)

.






Rxv is an M×(KM+KJ+K) matrix, C is a M×(KM+KJ+K) matrix and Rvv is an (KM+KJ+K)×(KM+KJ+K) matrix. If all of the second order statistics are available, then a solution to the problem would be to compute:

C=RxvRvv−1.  (eq. 46)

Wherein, the Rxv matrix includes components of Rxs and Rxu. Rxs is the cross correlation between the receiver data and the reference signal as defined above. An example of such a correlator is depicted as Xcorrelator 132 in FIG. 13. The channel matrix C above depends on Rxv and Rvv, thus it depends on the cross correlator Rxs (e.g., 132 in FIG. 13) through Rxv and through Rvv it depends on both the raw data x(n) and the training sequence s(n), since Rvv requires the training sequence autocorrelation Rss and the estimation of the whitened data u, which also depends on x(n).


However, a causality dilemma exists with regards to the temporally whitened noise process u(n) . In order to apply the autoregressive (AR) filter B(k) to x(n) to get u(n) in Equation 43 it is necessary to know both B(k) and H(k) . This can be accomplished using a bootstrapping technique that provides initial estimates for these quantities. The bootstrapping technique employs the assumption that the cross correlation terms between s(n) and u(n) are zero, Rsu(k)=0. This immediately decouples the estimation of H(k) from B(k) and thus the H(k) can be determined by direct correlation of the training sequences s(n) with x(n) .


Given H(k) the signal component excited by the s(n) can be subtracted out of x(n) , leaving i(n). The latter can be temporarily whitened using standard blind linear system identification techniques as described by Orfandis, independently applied to each antenna feed.


Once initial channel matrices H(k) and B(k) are obtained, the inverse filter in Equation 43 can be applied to obtain the u(n) process. The normal equations in Equation 46 can then be solved using fast recursive techniques for block Toeplitz matrices. This in turn permits improved estimation of the channel matrices.


Certain embodiments use ML estimation to obtain R:

{circumflex over (R)}uu=<u(n)u(n)H>n.


Once the channel estimates are obtained, they can be used to create models for A and Rii in the frequency domain for any subcarrier of an OFDM system. For frequency subcarrier f the channels can be written as:











A


(
f
)


=




k
=
0

K








H


(
k
)




e


-
2






π





kf












B


(
f
)


=




k
=
0

K








B


(
k
)




e


-
2






π





kf













R
ii



(
f
)


=


NB


(
f
)




R
uu




B
H



(
f
)













R
ii

-
1




(
f
)


=


1
N




D
H



(
f
)




R
uu

-
1




D


(
f
)




,





(

eq
.




47

)








where N is the length of the discrete Fourier transform and D(f)≡−B−1(f). All of the frequency components can be found using fast Fourier transforms (FFT). If an AR model is used for i(n), D(f) can be computed directly, making it possible to find the inverse covariance for all fusing a single matrix inverse.


Observation of the magnitudes of the elements of H(k) facilitates the “thresholding” of these values to obtain an estimate of the time support (number of nonzero samples) and an indication of which users are transmitting. For the latter, the peak lag (the largest magnitude of H(k)) can be used as a screen to determine which emitters are actually transmitting during a given transmission block or OFDM symbol, since each training sequence will be associated with a unique emitter.


A high level block diagram of the channel estimation process is shown in FIG. 13. In the simplest case, it is assumed that the training sequence 134 is uncorrelated with other training sequences and lagged training sequences and is further uncorrelated with the interference process, the channel estimation 136 reduces to nothing more than a cross correlation of the received data 130 against the training sequence. More exacting models attempt to remove the interference process and/or decorrelate against other emitters or channel delayed versions of the emitter waveforms. After the time domain impulse response is estimated, the frequency domain response can be computed using the FFT 138, from which is obtained the second order cross correlation statistics used in the standard normal equations for computing the beam-forming weights (shown at 139).


Bootstrapping H(k) and B(k).


Certain embodiments employ a bootstrapping procedure that may be summarized in the following steps:

    • 1. Solve H=RxSRSS−1 where:










H


(


H


(
0
)


,

H


(
1
)


,








H


(
K
)




)









R
xS



(



R
xs



(
0
)


,


R
xs



(
1
)


,









R
xs



(
K
)




)









R
SS




(





R
ss



(
0
)






R
ss



(
1
)









R
ss



(
K
)








R
ss



(

-
1

)






R
ss



(
0
)









R
ss



(

K
-
1

)






















R
ss



(

-
K

)






R
ss



(


-
K

+
1

)









R
ss



(
0
)





)

.






(

eq
.




48

)









    • 2. Cancel out s(n) by computing:










i


(
n
)


=


x


(
n
)


-




k
=
0

K








H


(
k
)





s


(

n
-
k

)


.










    • 3. Solve for B(n) using linear system identification techniques from the second order statistics of i(n) . One approach assumes a simple AR model and uses standard techniques for obtaining the AR model parameters. In this situation the inverse filter may be determined such that











-




k
=
0

K








D


(
k
)




i


(

n
-
k

)





=

u


(
n
)



,





where D(0)=−I . The Yule Walker equations for this problem yield:









R
jj



(
q
)


=




k
=
1

K









R
jj



(

q
-
k

)




D


(
k
)





,





where Rjj(q)≡<i(n+q)i(n)H>n. In matrix form this can be written as:

rJ=RJJD
RJJ−1rJ=D,

where







r
J




(



R
jj
H



(
0
)


,


R
jj
H



(
1
)


,





,


R
jj
H



(
K
)



)

H








R
jj



(






R
jj



(

-
1

)


,





R
jj



(

-
2

)









R
jj



(

-
K

)








R
jj



(
0
)






R
jj



(

-
1

)









R
jj



(

1
-
K

)






















R
jj



(

K
-
2

)






R
jj



(

K
-
3

)









R
jj



(

-
1

)





)







D




(



D
H



(
1
)


,


D
H



(
2
)


,





,


D
H



(
K
)



)

H

.






Once D(k) is obtained, an estimate for B(k) can be obtained by using the inverse filter to generate u(n), or by exciting the filter with a unit impulse and then obtaining the first K terms of the impulse response. u(n) can also be used directly in the estimation procedure in Equation 46.

    • 4. Use B(k) (or D(k)) to generate whitened u(k) to start the main algorithm.


In certain embodiment, only the bootstrap method is used to model the interference process and to train the FIR filter responses H(k) . For the frequency domain modeling B(f)=−D−1(f).


Multiple Transmit Antennas


A channel estimation procedure according to certain aspects of the invention can be extended to handle the case where the emitters have multiple antennas. One receiver model can be extended such that:








x


(
n
)


=





m
=
1

M










j
=
1

J









h
jm



(
n
)


*


s
jm



(
n
)





+

i


(
n
)




,





where hjm(n) is the channel seen from the m′th transmit antenna and the j′th emitter and sjm(n) is the complex information symbol transmitted from emitter j out of antenna m at time sample n. It becomes apparent that the m and j indices play the same role as the j indices played in the prior analysis. A simple Cartesian product mapping (j, m)→j′ to a new set of indices j′ yields the equation:







x


(
n
)


=






j


=
1


J











h

j





(
n
)


*


s

j





(
n
)




+


i


(
n
)


.






Since this is essentially the same framework as provided in the previous analysis, all of the previous results of the previous framework can apply. If knowledge of the full channel matrices is required, then this approach puts a stronger limit on how many emitters can be processed, due to both computational constraints and due to array loading constraints, depending on the number of samples or time bandwidth product (TBP) collected.


In most cases of interest, it can be assumed that a fixed set of transmit weights will be used at each transmitter and channel estimates can be performed by treating those weights as though they were part of the channel. This then provides a starting point of the analysis in Equation 42.


Data Covariance Approach


In certain embodiments the received data process may be modeled directly as a colored Gaussian noise process without attempting to estimate the interference noise process. This approach may provide the advantage in that it does not require coupling the estimation of the interference process to the estimation of the individual channels from each emitter. The received data vector may be modeled directly in the same way that the interference is modeled, using either the MA model:










x


(
n
)


=




k
=
0

K









B
x



(
k
)





u
x



(

n
-
k

)








(

eq
.




49

)








or the easier to estimate AR model:








-




k
=
0

K









D
x



(
k
)




x


(

n
-
k

)





=


u
x



(
n
)



,





where Dx(0)=−I . As in the interference case, the AR mode coefficients can be solved directly from the Yule Walker equations:









R
xx



(
q
)


=




k
=
1

K









R
xx



(

q
-
k

)




D


(
k
)





,





where

Rxx(q)≡<x(n+q)x(n)H>n.  (eq. 50)

In matrix form:

rX=RXXD
RXX−1rX=D,  (eq. 51)

where






r



(



R
xx
H



(
0
)


,


R
xx
H



(
1
)


,





,


R
xx
H



(
K
)



)

H








R
xx



(






R
xx



(

-
1

)


,





R
xx



(

-
2

)









R
xx



(

-
K

)








R
xx



(
0
)






R
xx



(

-
1

)









R
xx



(

1
-
K

)






















R
xx



(

K
-
2

)






R
xx



(

K
-
3

)









R
xx



(

-
1

)





)







D




(



D
H



(
1
)


,


D
H



(
2
)


,





,


D
H



(
K
)



)

H

.






Once the data process is estimated in the time domain, the frequency domain version can be obtained by noting the following relationships:












D


(
f
)




x


(
f
)



=

-


u
k



(
f
)












R
xx



(
f
)


=



ND


(
f
)



-
1




R


u
x



u
x






D


(
f
)



-
H













R
xx

-
1




(
f
)


=


1
N




D
H



(
f
)




R


u
x



u
x



-
1




D


(
f
)




,





(

eq
.




52

)








where Ruxux is the spatial covariance matrix of the ux(n) process, which can be easily estimated using sample averages at the output of the estimated filter D(k) . This result is particularly appealing because it demonstrates that the inverse covariance matrix can be obtained for all frequency bins using only one inverse of Ruxux.


This approach can be further extended by applying the same process to bandlimited signals at the output of a band pass filter. Thus independent sets of signals can be considered at the output of multiple band pass filters. If there are K band pass filters then there would be needed only K matrix inversions of Ruxux(k) , the covariance matrix of the whitened noise process associated with sub-band k.


A data AR model estimator is illustrated in FIG. 14. Delayed versions 142 of the received data vectors 140 are cross correlated 141 against the received data 140 in order to set up the Yule Walker equations. Those equations may be solved for the whitening model 143 which must be applied to the original data to determine Ruxux in 144. In one embodiment, process 144 may obtain the whitened data autocovariance described above in Equation 49. In another embodiment, a similar interference covariance is used in Equation 44 and the definition of Ruxux above. The inverse data covariance matrix is obtained in the frequency domain after an FFT 145 is performed on the whitening matrix coefficients. The inverse covariance matrix 146 as a function of frequency can then be used to obtain the beam-forming weights at the receiver. This corresponds to the data covariance approach that gives the inverse matrix as Equation 52. It is noted that Equation 52 is a powerful result that demonstrates that only one matrix inversion needs to be performed to obtain all matrix inversions as a function of frequency.


Non Causal Filters


In certain embodiments, the implementation of a “localized” non-causal model for the temporal filters is of interest because of the use of a cyclic prefix in most OFDM waveforms. The cyclic prefix can create channels that appear to have non-causal impulse responses. Indeed a bandlimited unit impulse will have energy at negative time values prior to transmission, once the cyclic prefix is added. The received data model can be examined and extended to handle this situation. Thus, from Equation 49:








x


(
n
)


=






k
=

-

K
1




K
2










B
x



(
k
)





u
x



(

n
-
k

)




-




k
=

-

K
1




K
2










D
x



(
k
)




x


(

n
-
k

)





=


u
x



(
n
)




,







x


(
n
)


-




k

0










D
x



(
k
)




x


(

n
-
k

)





=


u
x



(
n
)








where Dx(0)=−I . Minimizing the variance of ux(n), and using Equation 50, the Yule Walker equations are:











R
xx



(
q
)


=





k
=

-

K
1




k

0



K
2










D
x



(
k
)






R
xx



(

q
-
k

)


.







(

eq
.




53

)








It is apparent then, that these relations hold independent of the ux(n) spatial coloring or non identity value of Ruxux(k) . Defining:







r
x




(



R
xx
H



(

-

K
1


)


,


R
xx
H



(

1
-

K
1


)


,





,


R
xx
H



(

K
2

)



)

H








R
xx



(






R
xx



(
0
)


,





R
xx



(

-
1

)









R
xx



(


-

K
1


-

K
2


)








R
xx



(
1
)






R
xx



(
0
)









R
xx



(

1
-

K
1

-

K
2


)






















R
xx



(


K
1

+

K
2


)






R
xx



(


K
1

+

K
2

-
1

)









R
xx



(
0
)





)







D




(



D
T



(

-

K
1


)


,


D
T



(

1
-

K
1


)


,





,


D
T



(

-
1

)


,


D
T



(
1
)


,





,


D
T



(

K
2

)



)

T

.






It will be appreciated that the entries corresponding to the zero index are omitted. The equations can then be solved from Equation 51. A similar extension to the direct estimation of non-causal channel responses can be employed.


Adopting the multi-user non-causal channel model from Equation 42:








i


(
n
)


=


x


(
n
)


-




k
=

-

K
1




K
2









H


(
k
)




s


(

n
-
k

)






,





the Yule-Walker equations can then be written as:


















R
xs



(
k
)


=




m
=

-

K
1




K
2









H


(
m
)





R
ss



(

k
-
m

)















H


(


H


(

-

K
1


)


,

H


(

1
-

K
1


)


,








H


(

K
2

)




)













R
xS



(



R
xs



(

-

K
1


)


,


R
xs



(

1
-

K
1


)


,









R
xs



(

K
2

)




)










R
SS








(






R
ss



(
0
)


,





R
ss



(
1
)









R
ss



(


K
1

+

K
2


)








R
ss



(

-
1

)






R
ss



(
0
)









R
ss



(


K
1

+

K
2

-
1

)






















R
ss



(


-

K
1


-

K
2


)






R
ss



(


-

K
1


-

K
2

+
1

)









R
ss



(
0
)





)

.







(

eq
.




54

)








The equations are solved by H=RxSRSS−1.


Frequency Domain Versions


In certain embodiments, the Yule Walker Equations can be solved in the frequency domain by noting that a convolution is just a multiply in the frequency domain. However, FFTs of different sizes and/or decimation may be required because the cross correlations and the convolutions with the channels can have different time support. For example, by taking the DFT of both sides of Equation 54:

{circumflex over (R)}xs(e)=Ĥ(e){circumflex over (R)}ss(e),  (eq. 55)

where {circumflex over (·)} the notation is used to indicate the transform. In this example, it is assumed that a truncated transform for









H
^



(

e

j





ω


)


=




k
=

-

K
1




K
2





H


(
k
)




e


-
jk






ω





,





as though the support for H(k) was confined to [−K1, K2]. Using the Discrete Fourier Transform yields:

{circumflex over (R)}xs(e)={circumflex over (X)}(e)ŝH(e)
{circumflex over (R)}ss(e)=ŝ(e)ŝH(e)
{circumflex over (X)}(e)=Ĥ(e)ŝ(e).  (eq. 56)

The latter equation is undetermined unless constraints are put on Ĥ(e), (e.g. finite time support). The easiest way to handle a finite time support constraint in the frequency domain is to solve Equation 55 in the frequency domain using a DFT of length K1+K2+1. Then:

Ĥ(e)={circumflex over (R)}xs(e){circumflex over (R)}ss−1(e)

can be set for ω=2πq/ N, q=−K1. . . K2 . This is computationally cheaper than solving Equation 54 directly in the time domain. The simplifications in Equation 56 typically cannot be applied here because the DFT is truncated. It is also possible to use FFT techniques to solve the received data modeling problem by taking the DFT of both sides of Equation 53.


Precomputations and Single Emitter Analysis


In certain embodiments, certain computations related to the computation of H(k) can be computed in advance of any received data and stored in memory. To further simplify the analysis the assumption of computing the channel for a single emitter can be considered, because this assumption is likely to be sufficient to obtain a near optimal linear beam-former. The latter assertion arises from the fact that this is true in the absence of multipath. The joint steering vector estimate is not needed to obtain optimal beam-forming weights.


Assuming a single emitter, then the cross correlation statistic may be written:














R
xs



(
k
)










x


(
n
)





s
H



(

n
-
k

)





n








=




1
N


X






D
k


s


,







(

eq
.




57

)








where X is the M×N received (non-conjugated) data matrix. The n′th column of X is the data vector over M antennas received at time sample n. Dk is the circular shift operator corresponding to a delay or right shift of k samples and s is the N×1 conjugated training signal whose n′th element is s*(n). Defining

S≡[D−K1s, D1−K1s, . . . , s, D1s, D2s, . . . DK2s],

the channel estimator becomes:

H=XS(SHS)−1.

Typically, the matrix S(SHS)−1 can be precomputed. A straightforward extension to the multiple user case can also be made and the channel estimating matrix may also be precomputed in that case.


Waveform Design


Certain embodiments address a number of design parameters when determining what training waveforms should be used to obtain channel estimates. Parameters that are addressed include constraints imposed by standards such as WiMAX, and the efficiency of the estimation procedure. In one embodiment, design criteria include

    • 1. reducing the peak to average power ratio (PAPR) of the transmitted waveform,
    • 2. enforcing minimal correlation between training sequences from different emitters,
    • 3. enforcing minimal correlation between the training sequence and delayed or lagged versions of the training sequence,
    • 4. enforcing minimal correlation between the training sequence and delayed versions of other training sequences,
    • 5. enabling fast transform techniques so that many cross correlations can be performed simultaneously (e.g. Hadamard sequences),
    • 6. restricting training sequences to the subbands that are being used by the emitter, and
    • 7. enforcing stacked carrier spreading so that the training sequences have the same frequency spreading code as their bearer traffic waveforms.


In certain circumstances, some of these criteria and goals may conflict with other criteria and goals. In certain embodiments, criteria are weighted according to the design priorities of the system. Each goal may offer a mathematical and in some cases linear constraint and may be associated with a mathematical objective function. The constraints and objectives can be summed together with an appropriate weighting. Thus the design of these waveforms can be obtained using parameter optimization. One approach for doing this is to deploy stochastic global parameter optimization techniques.


Time Domain Preamble Symbol Design


Certain embodiments provide systems compliant with promulgated standards. In one example, consistent with requirements of the WiMAX standard, one possible time domain preamble symbol 150 comprises a pseudo-noise (PN) sequence having a length of 512 as illustrated in FIG. 15. To support downlink messaging, each base station (BS) uses a unique PN sequence. In order to avoid the “out-band” interference, the “chip” rate of the PN sequence should be less than 768. In the example, 576 chips are used during 0.1028 ms, where the first 64 chips 151correspond to the cyclic prefix period 153 and the following 512 chips 154 correspond to the OFDM symbol 155. The first 64 chips are determined by the cyclic prefix.


Synchronization Procedure with Time Domain Preamble


Certain embodiments provide a mobile station (MS) that stores a sufficient number of PN sequences to cover those used by interfering base stations. Each BS can use a unique PN sequence. Initially, the MS searches for the strongest PN sequence in order to obtain frame timing information. When the MS obtains the frame timing information and decodes control channel information successfully, the MS can setup a connection with this BS. During the initial synchronization search, the MS can not only search the strongest PN sequence, but also can search the second “strongest” and third “strongest” paths. After the connection between the MS and BS has been established, the BS may provide the PN sequences used by the neighboring cells to the MS. The number of PN sequences, the PN generator and the neighboring cell messages are typically chosen according to the network geometry and by the criterion suggested earlier.


Channel Interpolation and Extrapolation in Time Varying Channels


Certain embodiments extend the technique to handle time varying channels by proposing various schemes that permit the interpolation of the channel from one time block to the next. This is readily accomplished by further parameterizing the channel matrices, Ruu, B(k), H(k) and D(k) by a continuous time offset t. It can be assumed that the channel is sampled at discrete times tq. The channel is assumed to be quasi-stationary during the training burst at time tq. Possible interpolation or extrapolation schemes include:

    • 1. linear interpolation,
    • 2. log linear interpolation (linear interpolate the logarithm of the channel quantities),
    • 3. polynomial and log polynomial interpolation, and
    • 4. Spline and log spline interpolation.


Certain embodiments use model based interpolation, moving specular reflectors and their known effects on the channel are modeled. Each of these types of interpolation and extrapolation can also be performed as needed in the frequency domain.


Simplifications and Additional Embodiments


Certain embodiments employ one or more simplifications of the techniques described above. Some simplifications allow a reduction in hardware complexity. Also channel models used as downlink channels can be deployed for use in uplink applications. In one example, time division duplex systems (TDD) can exploit channel reciprocity including the use of receiver weights as transmit weights. Thus knowledge of the channel as a function of frequency can greatly facilitate linear beam-forming/stacked carrier spreading at both ends of the link. Uplink processing may proceed using the same basic signal processing structure.


Many simplifications can be considered in the area of reducing the complexity of channel estimation. In certain embodiments, it may be assumed that Rsu(k)=0, so that the estimation of the individual channels, H(k), and the estimation of the interference statistics B(k) or D(k) would decouple. Further simplifications along these lines assume that the training signals sj(n) are orthonormal or even that they are white, that is orthogonal to shifts, sj(n−k) ⊥sq(n) . This would completely diagonalize Rss in Equation 48 making the solution of the Yule Walker equations trivial, requiring only cross correlations with sj(n−k).


In certain embodiments, Hadamard codes can be used for sq(n), making the correlation over multiple emitters (q), much more rapid, whilst maintaining orthogonality. In certain embodiments, fast convolution algorithms may be employed to quickly obtain the correlations over multiple lags. To further simplify processing, certain embodiments also attempt to learn the channels of all the nearby emitters and then treat the interference process as white noise, thereby eliminating the need for certain estimations.


Additional Descriptions of Certain Aspects of the Invention


The foregoing descriptions of the invention are intended to be illustrative and not limiting. For example, those skilled in the art will appreciate that the invention can be practiced with various combinations of the functionalities and capabilities described above, and can include fewer or additional components than described above. Certain additional aspects and features of the invention are further set forth below, and can be obtained using the functionalities and components described in more detail above, as will be appreciated by those skilled in the art after being taught by the present disclosure.


With reference to FIGS. 7, 8, 12 and 19, certain embodiments of the invention provide methods and systems for enhancing interference cancellation. In some of these embodiments the methods comprise receiving a signal 1900 including stacked carriers 80 at a receiver 121, obtaining, at 1908, samples 82 from two or more of the stacked carriers 80, creating a data vector from the samples 1910, wherein the two or more stacked carriers are selected using a steering vector received during a synchronization 1901 of the receiver. In some of these embodiments, spacing between the stacked carriers is variable. In some of these embodiments, the steering vector is calculated to obtain cancellation of interference from another receiver. In some of these embodiments, the steering vector is calculated based on a time domain channel estimation. Some of these embodiments obtain frequency domain interference cancellation.


Certain embodiments of the invention provide specialized time domain training sequences 1901 for use in channel estimation at 1902 adhering to one or more design criterion. In some of these embodiments, global stochastic parameter estimation techniques are used to facilitate the design of the training sequences. In some of these embodiments, time domain training sequences that perform simple cross correlation are used to obtain a channel estimate for use in stacked carrier beamforming and/or for use in OFDM based spatial beamforming at 1906.


Certain embodiments of the invention provide systems and methods that use time domain training sequences to solve Yule Walker equations. In some of these embodiments, estimation of the interference noise process is provided. In some of these embodiments, bootstrapping techniques enable a first attempt to estimate the interference process and remove it from the environment, prior to solving the joint Yule Walker equations. In some of these embodiments, direct data process estimation is used including an inverse and/or an AR model.


In some of these embodiments, at least one of fast convolution and FFT techniques facilitate an economical solution of the Yule Walker equations and/or multiple cross correlations. In some of these embodiments, data process AR model facilitate direct computation of an inverse covariance matrix, in the frequency domain over some or many subcarriers, without having to perform additional matrix inversions. In some of these embodiments, an inverse interference process AR model is used to directly compute the inverse interference covariance matrix, in the frequency domain over one or more subcarriers, without having to perform additional matrix inversions.


In some of these embodiments, one or more of a data process and an interference process AR model is used to directly compute inverse covariance matrices for facilitating computation of beamforming weights at 1904 in FIG. 19 for a plurality of subcarriers to reduce the effects of frequency dispersion. In some of these embodiments, Multiple User Detection is used to further enhance receiver performance. In some of these embodiments, non-causal filters and non-causal filter estimation is used to support a cyclic prefix. In some of these embodiments, the cycle prefix is used in an OFDM system. In some of these embodiments, precomputation is used to obtain the inverses of multiple signal cross and auto correlations to facilitate solution of Yule-Walker equations.


Certain embodiments of the invention provide systems and methods for interpolating or extrapolating matrix channel estimates and whitening matrix filter estimates over time and frequency. Some of these embodiments include the use of one or more of linear interpolation, log linear interpolation, linear interpolation of the logarithm of channel quantities, polynomial interpolation, log polynomial interpolation, spline interpolation and log spline interpolation. In some of these embodiments, interpolation and extrapolation is used for maintaining linear beamforming weights as a function of time or frequency.


Certain embodiments of the invention provide systems and methods that use state space or filter based modeled interpolators in the frequency domain or time domain to interpolation of the channels, or whitening matrix-filters over time or frequency, for maintaining linear beamforming weights as a function of time or frequency. In some of these embodiments, a linear weight computation procedure models all important emitter channels, and treats unmodeled interference as white noise. In some of these embodiments, a linear weight computation procedure models one or more important emitter channels, and models either the interference process or the entire data process using inverse modeling or even direct modeling using standard system identification procedures. In some of these embodiments, a deinterleaver is used for facilitating the use of stacked carrier processing by forming vectors in stacks and clustering vectors that are adjacent in frequency. In some of these embodiments, the system is an IEEE 802.16 compliant system. In some of these embodiments, the system uses subcarrier permutations in repetition codes.


Certain embodiments of the invention provide systems and methods of interference cancellation comprising a stacked carrier for enhancing interference cancellation for a set of repetition codes spread over frequency subcarriers in an arbitrary manner. Some of these embodiments further comprise identification of interfering spread waveforms that have at least two carriers that overlap. In some of these embodiments, non-overlapping subcarriers are treated as zeros in the measured steering vectors. In some of these embodiments, a stacked carrier is used to enhance interference cancellation for large spreading factor and for exploiting sparse channels to affect a matrix inverse when computing linear beamforming/despreading weights. In some of these embodiments, backsubstitution is used to simplify processing wherein data is ordered such that the in-cell data is at the bottom of the vector.


In some of these embodiments, space time adaptive processing (STAP) is used to facilitate beamforming of communications signals for OFDM waveforms. In some of these embodiments, a processing chain is provided for estimating communication signals, comprising a collection of first and second order statistics, using a matrix fast filter estimation, using a bank of bandpass filters to support stacked carrier, and using FFTs to improve computational complexity to facilitate beamforming in the frequency domain for OFDM waveforms. In some of these embodiments, non-causal filters provide fast matrix valued adaptive space time processing for interference cancellation in OFDM and 802.16 (WiMax) systems.


Although the present invention has been described with reference to specific exemplary embodiments, it will be evident to one of ordinary skill in the art that various modifications and changes may be made to these embodiments without departing from the broader spirit and scope of the invention. Accordingly, the specification and drawings are to be regarded in an illustrative rather than a restrictive sense.


Although the present invention has been described with reference to specific exemplary embodiments, it will be evident to one of ordinary skill in the art that various modifications and changes may be made to these embodiments without departing from the broader spirit and scope of the invention. Accordingly, the specification and drawings are to be regarded in an illustrative rather than a restrictive sense. Throughout the foregoing description, for the purposes of explanation, numerous specific details were set forth in order to provide a thorough understanding of the invention. It will be apparent, however, to one skilled in the art that the invention may be practiced without some of these specific details. Embodiments may include various operations as set forth above or fewer or more operations, or operations in an order different from the order described.


Accordingly, the scope and spirit of the invention should be judged in terms of the claims which follow as well as the legal equivalents thereof.

Claims
  • 1. A wireless communications system for estimating performance of a communications channel with co-channel interference, the system comprising: a training sequence generator configured to produce a training sequence preamble in time domain; anda processor having a cross-correlator, a channel impulse response estimator, and a Fast Fourier Transform (FFT) calculator, wherein the cross-correlator cross-correlates raw data in a received signal in the communications channel with the training sequence preamble, the channel impulse response estimator estimates time domain impulse response for the received signal cross-correlated with the training sequence preamble, and the FFT calculator computes frequency domain response from the output of the channel impulse response estimator;wherein the processor uses second order cross-correlation statistics obtained from the output of the FFT calculator to create a channel matrix that estimates performance of the communications channel with co-channel interference.
  • 2. The system of claim 1, wherein the processor uses the channel matrix in computing beam-forming weights for a plurality of subcarriers for reducing co-channel interference in the communications channel.
  • 3. The system of claim 2, wherein the system further includes an interference whitening filter, which uses an autoregressive (AR) model and FFT to obtain an inverse covariance matrix in frequency domain that, along with the channel matrix, facilitates computation of the beam-forming weights for the plurality of subcarriers to reduce the effect of frequency dispersion.
  • 4. The system of claim 2, wherein characteristics of at least some of the co-channel interference is different from additive white Gaussian noise (AWGN) characteristics.
  • 5. The system of claim 2, where the received signal has code repetitions.
  • 6. The system of claim 5, wherein neighboring cells in the wireless communications system use same code repetition numbers and same data block size.
  • 7. The system of claim 6, wherein de-permutation is performed to extract information about number of code repetitions in the received signal.
  • 8. The system of claim 6, wherein the signal in the communications channel is received by a receiver having a plurality of antennas.
  • 9. The system of claim 8, wherein the number of antennas is factored in while exploiting code repetitions to cancel the co-channel interference.
  • 10. The system of claim 1, wherein the channel matrix is sparse.
  • 11. A method wireless communications system for estimating performance of a communications channel with co-channel interference in a wireless communications system, the method comprising: receiving a signal with raw data in a communications channel;producing a training sequence preamble in the time domain;cross-correlating, at a processor, the raw data in the received signal with the training sequence preamble, the processor having a cross-correlator, a channel impulse response estimator, and a Fast Fourier Transform (FFT) calculator;estimating, at the channel imulse response estimator, time domain impulse response for the received signal cross-correlated with the training sequence preamble, andperforming, at the FFT calculator, Fast Fourier Transform (FFT) on the output of the channel impulse response estimator to calculate frequency domain response;using second order cross-correlation statistics obtained from the FFT process to create a channel matrix that estimates performance of the communications channel with co-channel interference.
  • 12. The method of claim 11, wherein the method further includes: using the channel matrix to compute beam-forming weights for a plurality of subcarriers for reducing co-channel interference in the communications channel.
  • 13. The method of claim 12, wherein the method further includes: providing a whitening filter which uses an autoregressive (AR) model and FFT to obtain an inverse covariance matrix in frequency domain;using the inverse covariance matrix in frequency domain, along with the channel matrix, to compute the beam-forming weights for the plurality of subcarriers to reduce the effect of frequency dispersion.
  • 14. The method of claim 12, wherein characteristics of at least some of the co-channel interference is different from additive white Gaussian noise (AWGN) characteristics.
  • 15. The method of claim 12, where the received signal has code repetitions.
  • 16. The method of claim 15, wherein neighboring cells in the wireless communications system use same code repetition numbers and same data block size.
  • 17. The method of claim 16, wherein de-permutation is performed to extract information about number of code repetitions in the received signal.
  • 18. The method of claim 16, wherein the signal in the communications channel is received by a receiver having a plurality of antennas.
  • 19. The method of claim 18, wherein the number of antennas is factored in while exploiting code repetitions to cancel the co-channel interference.
  • 20. The method of claim 11, wherein the channel matrix is sparse.
CROSS-REFERENCE TO RELATED APPLICATIONS

The present application is a continuation of U.S. patent application Ser. No. 13/411,510 filed Mar. 2, 2012, entitled “Wireless Communications Systems and Methods,” now U.S. Pat. No. 9,088,446, which is a divisional application of U.S. patent application Ser. No. 12/276,261 filed Nov. 21, 2008, entitled “Systems and Methods for Channel Based Beamforming for Stacked Carrier Multiple Access,” now U.S. Pat. No. 8,300,742, which claims priority from U.S. Provisional Patent Application No. 60/989,802 filed Nov. 21, 2007, entitled “Systems and Methods for Channel Based Beamforming for Stacked Carrier Multiple Access,” which are incorporated by reference herein in their entirety.

US Referenced Citations (37)
Number Name Date Kind
5982327 Vook et al. Nov 1999 A
6128276 Agee Oct 2000 A
6512737 Agee Jan 2003 B1
6598014 Rabideau et al. Jul 2003 B1
6788661 Ylitalo et al. Sep 2004 B1
7269238 Lee et al. Sep 2007 B2
7327800 Oprea et al. Feb 2008 B2
7970081 Guey Jun 2011 B2
8064532 Maltsev et al. Nov 2011 B2
8761312 Stamoulis Jun 2014 B2
20020027957 Paulraj et al. Mar 2002 A1
20020176485 Hudson Nov 2002 A1
20040001554 Sun Jan 2004 A1
20040174940 Kim Sep 2004 A1
20050111538 Wernaers May 2005 A1
20050276355 Chow Dec 2005 A1
20060023799 Kang et al. Feb 2006 A1
20060274841 Wang et al. Dec 2006 A1
20060294170 Matsuoka et al. Dec 2006 A1
20070041475 Koshy et al. Feb 2007 A1
20070195974 Li et al. Aug 2007 A1
20070201349 McCoy Aug 2007 A1
20070211809 Kim Sep 2007 A1
20070280367 Nakao et al. Dec 2007 A1
20070281640 Coupain et al. Dec 2007 A1
20080037691 Papathanasiou Feb 2008 A1
20080076477 Hedayat Mar 2008 A1
20080089396 Zhang et al. Apr 2008 A1
20080144730 Akella et al. Jun 2008 A1
20080273639 Mikhael et al. Nov 2008 A1
20090052514 Sampath Feb 2009 A1
20110026578 Chen et al. Feb 2011 A1
20110182241 Hansen et al. Jul 2011 A1
20110317683 Branlund et al. Dec 2011 A1
20120033618 Wallace et al. Feb 2012 A1
20120057080 Kim et al. Mar 2012 A1
20130089052 Kim Apr 2013 A1
Non-Patent Literature Citations (10)
Entry
Bromberg, et al., The Use of Programmable DSPs in Antenna Array Processing, Oct. 2001, pp. 57-95, John Wiley and Sons Ltd.
Golub, et al., Matrix Computations, 1996, The John Hopkins University Press, Baltimore Maryland.
Haykin, Simon, Adaptive Filter Theory, 2002, Prentice Hall Inc., Upper Saddle River, New Jersey.
Maryak, et al., Global Optimization via SPSA, 1999, The John Hopkins University Applied Physics Laboratory, Laurel, Maryland.
Myers, Douglas, G., Digital Signal Processing, Efficient Convolution and Fourier Transform Techniques, 1990, pp. 85-112, Prentice Hall of Australia Pty. Ltd.
Orfanidis, S.J., Optimum Signal Processing: An Introduction, 1988, McGraw Hill Publishing, New York, New York.
Struckman, et al., A Multidimensional Bayesian Global Search Method Which Incorporates Knowledge of an Upper Bound, University of Louisville, 1989, pp. 591-596.
Thirumalai, S., High Performance Algorithms to Solve Toeplitz and Block Toeplitz Matrices, 1996, Ph.D. Thesis—University of Illinois, Urbana-Champaign.
Verdu, S., Multiuser Detection, 1998, Cambridge University Press, New York, New York.
Yoshii, et al.,“Interference Mitigation with Coordinated Symbol Repetition”, IEEE, S802.16M—Jul. 2011, Nov. 7, 2007, pp. 1-7.
Related Publications (1)
Number Date Country
20150381259 A1 Dec 2015 US
Provisional Applications (1)
Number Date Country
60989802 Nov 2007 US
Divisions (1)
Number Date Country
Parent 12276261 Nov 2008 US
Child 13411510 US
Continuations (1)
Number Date Country
Parent 13411510 Mar 2012 US
Child 14804094 US