WO2010031131A1

WO2010031131A1 - Vector quantization in wireless communication

Info

Publication number: WO2010031131A1
Application number: PCT/AU2009/001238
Authority: WO
Inventors: Iain Bruce Collings; Daniel Ryan; Jean-Marc Valin
Original assignee: Commonwealth Scientific and Industrial Research Organization CSIRO
Current assignee: Commonwealth Scientific and Industrial Research Organization CSIRO
Priority date: 2008-09-18
Filing date: 2009-09-18
Publication date: 2010-03-25
Anticipated expiration: 2011-03-18
Also published as: EP2327170A1; US20110044400A1; EP2327170A4; AU2009295275A1; US8396163B2

Abstract

Disclosed is a method of communicating a complex vector, using one or more index bits, the method comprising quantizing the complex vector using a codebook, the codebook comprising a plurality of complex vectors mapped from real vectors that lie on a reflected simplex where the simplex is of dimension one less than twice the length of the complex vector; indexing the quantized vector to form the one or more index bits; and transmitting the index bits to the transmitter, thereby communicating the complex vector.

Description

VECTOR QUANTIZATION IN WIRELESS COMMUNICATION

Technical Field

The present invention relates generally to wireless communication and, in particular, to vector quantization for limited feedback from receiver to transmitter.

Background

Multiple-input multiple-output (MIMO) systems can provide increased reliability in wireless communication links by exploiting the spatial diversity due to the increased number of transmit-receive paths. A simple technique to obtain the highest possible diversity order is to employ transmit beamforming and receive combining, which also improves the array gain. This technique requires that the transmitter has channel state information in the form of a transmit beamforming vector. It is often impractical to have a full reciprocal channel from receiver to transmitter to enable the transmitter to estimate the forward channel state information. Instead, the receiver estimates the channel state information, computes the corresponding beamforming vector, and encodes the beamforming vector in a small number of bits. These bits are sent via a feedback path to enable the transmitter to generate the beamforming vector. Such systems are known as limited feedback systems.

The most straightforward approach to designing a limited feedback system is to employ scalar quantization, where each component of the beamforming vector is quantized and encoded separately. In more advanced limited feedback systems, the transmitter and receiver share a codebook of possible beamforming vectors indexed by a number of bits. The receiver chooses a beamforming vector from the codebook on the basis of maximizing the effective signal-to-noise ratio (SNR) after combining, and sends the corresponding bits to the transmitter. Beamforming vector codebooks are conventionally designed using the minimum number of feedback bits possible for a given effective SNR after combining, i.e. neglecting the search and storage requirements for the codebook. Codebook design strategies generally use numerical optimization techniques, or for larger systems, the codebooks can be randomly generated (i.e. random vector quantization, or RVQ). Such random codebooks have been shown to be asymptotically optimal as the number of bits and transmit antennas increase.

Unfortunately, the codebook size increases exponentially with the number of transmit antennas to maintain a given effective SNR or capacity loss with respect to the optimal unquantized system. Since RVQ codebooks have no structure, an exhaustive search is usually required to find the bits encoding a given beamforming vector, or vice versa. For time-varying channels, the resulting delay due to the excessive search time reduces the effectiveness of the beamforming vector when employed at the transmitter. The computation required for such an exhaustive search also consumes power, which is undesirable for low-power mobile wireless devices. Non-exhaustive methods for searching unstructured codebooks at the expense of increased memory requirements have been documented. One of these methods is a tree-search, where storage of the tree and codebook is required. An additional consequence of the exponential growth in codebook size with antenna number is that storage of the codebook may be infeasible for large numbers of antennas.

The problem of codebook search time and storage requirements is of particular importance to multiuser systems, where quantization errors increase the interference between users.

Summary Disclosed are arrangements which seek to ameliorate the disadvantages of existing methods by using Reflected Simplex codebooks in beamforming vector quantization for limited feedback MIMO systems. The disclosed Reflected Simplex codebooks adhere to a geometric construction, consisting of integer coordinates that lie within the original and reflected images about the axes of a (2Nχ - l)-dimensional simplex, where Nj is the number of transmit antennas. Also disclosed is a method of indexing the Reflected Simplex codebook with low time and storage requirements. Also disclosed is an efficient method of searching through the codebook for the optimal quantized beamforming vector. According to a first aspect of the present invention, there is provided a method of communicating a complex vector, using one or more index bits, the method comprising: quantizing the complex vector using a codebook, the codebook comprising a plurality of complex vectors mapped from real vectors that lie on a reflected simplex where the simplex is of dimension one less than twice the length of the complex vector; indexing the quantized vector to form the one or more index bits; and transmitting the index bits to the transmitter, thereby communicating the complex vector.

According to a second aspect of the present invention, there is provided a method of wireless communication between a transmitter comprising a plurality of transmit antennas and a receiver over a channel, the method comprising: estimating channel state information for the channel; computing an optimal beamforming vector for the plurality of transmit antennas from the channel state information; quantizing the optimal beamforming vector using a codebook, the codebook comprising a plurality of complex vectors mapped from real vectors that lie on a reflected simplex where the simplex is of dimension one less than twice the length of the optimal beamforming vector; indexing the quantized vector to form one or more index bits; and transmitting the index bits to the transmitter. According to a third aspect of the present invention, there is provided a receiver for implementing the method according to the second aspect.

According to a fourth aspect of the present invention, there is provided a system for wireless communication over a channel, the system comprising: a transmitter comprising a plurality of transmit antennas; and a receiver adapted to: estimate channel state information for the channel; compute an optimal beamforming vector for the plurality of transmit antennas from the channel state information; quantize the optimal beamforming vector using a codebook, the codebook comprising a plurality of complex vectors mapped from real vectors that lie on a reflected simplex where the simplex is of dimension one less than twice the length of the optimal beamforming vector; index the quantized vector to form one or more index bits; and transmit the index bits to the transmitter.

Other aspects of the invention are also disclosed.

Brief Description of the Drawings Embodiments of the present invention are described hereinafter with reference to the drawings, in which:

Fig. 1 is a block diagram showing a limited feedback MIMO wireless communication system on which the embodiments of the present invention may be practised; Fig. 2 is a plot of the vectors in the PVQ codebook D(I, 2);

Fig. 3 is a flowchart illustrating a method of limited feedback wireless communication carried out by the system of Fig. 1 ;

Fig. 4 is a flowchart illustrating a method of obtaining an index for a complex vector in the Reflected Simplex codebook, as used in the method of Fig. 3; Fig. 5 is a plot illustrating the zigzag ordering of complex values used in the indexing method of Fig. 4;

Fig. 6 is a flowchart illustrating a method of searching the Reflected Simplex codebook, as used in the method of Fig. 7; and

Fig. 7 is a flowchart illustrating a method of encoding a real vector to index bits, as used in the method of Fig. 3.

Detailed Description

Where reference is made in any one or more of the accompanying drawings to steps and/or features, which have the same reference numerals, those steps and/or features have for the purposes of this description the same function(s) or operation(s), unless the contrary intention appears.

Fig. 1 is a block diagram showing a limited feedback MIMO wireless communication system 100 on which the embodiments of the present invention may be practised. The system 100 comprises a transmitter 110, which transmits data symbols wirelessly over a narrow-band channel 125 through Nj transmit antennas 120-1 to 120-iVτ-

The data symbols are received by a receiver 130 through N_R receive antennas 140-1 to

140-N_R. The system 100 also comprises a low-rate feedback link 150 (indicated by a broken line) through which the receiver sends state information about the channel 125 to the transmitter 110. The transmitter 110 uses the state information to maximise the performance of the transmission to the receiver 130.

Since the channel 125 is narrow-band, a complex transmitted symbol x is transformed to a complex received symbol y as follows:

>> = z^rHwx + z^rn, (1) where w is the unit magnitude complex beamforming vector (of length Nj) applied at the transmitter 110, H is a JVR-by-JVτ complex- valued matrix representing the effect of the channel 125, z is the unit magnitude complex combining vector (of length NR) applied at the receiver 130, and n is the complex noise vector (of length N_R) at the receiver 130. Assuming the noise is independent, identically distributed (i.i.d), circularly symmetric complex Gaussian with variance N₀, the signal-to-noise ratio (SΝR) p at the receiver 130 is given by:

where E_x is the average transmitted symbol power and

Γ(H) = |z^THw|² (3) is the effective channel gain. Given w and H, the (unit magnitude) combining vector z that maximises the received SΝR is given by the Maximum Ratio Combining (MRC) formula:

From equation (3), the resulting effective channel gain (the MRC gain) is:

The optimal infinite-precision (unquantized) beamforming vector w_opt that maximises the MRC gain is given by the right-singular unit vector of H corresponding to the largest singular value of H. The beamforming scheme utilising w_opt is known as maximum ratio transmission (MRT). The effective channel gain when using MRT and MRC is denoted F_MRT,MRC(H) and is given by the square of the largest singular value of H. In the limited feedback system 100 of Fig. 1, the receiver 130 and the transmitter

110 share a codebook C of possible beamforming vectors, which are indexed by ceil(log₂(|C|)) bits. The optimal quantized beamforming vector w_opt in the codebook C may be computed as:

The denominator is present in equation (6) to ensure the average transmitted signal power E_x is constant.

Because in general w_opt ≠ w_opt , the effective channel gain using w_opt (computable using equation (5)) is less than the MRC-MRT gain Γ_MRT,_MRC(H), by an amount depending on the granularity or quantization step size of the codebook C. The design of the codebook C is therefore critical for the performance of the system 100.

Fig. 3 is a flowchart illustrating a method 300 of limited feedback wireless communication carried out by the system 100 of Fig. 1. The steps on the left of the dotted line 305 are carried out by the receiver 130, while the steps on the right are carried out by the transmitter 110. The method 300 starts at the step 310, where the receiver 130 estimates the channel matrix H using known methods, for example from the characteristics of a known training sequence transmitted by the transmitter 110. In step 320, the receiver 130 computes the optimal (unquantized) beamforming vector w_opt. This is done using the singular value decomposition of the channel matrix H as described above. In the next step 330, the receiver 130 finds the index bits encoding the optimal beamforming vector w_opt, as described in more detail below. Step 330 involves quantization of w_opt and indexing of the quantized vector. Next, the receiver 130 in step 340 transmits the index bits to the transmitter 110 (indicated in Fig. 3 by the broken line arrow from the receiver side to the transmitter side) over the feedback link 150 of Fig. 1 (which is assumed to be error-free). At step 350 the receiver 130 computes the (unit magnitude) MRC combining vector Z_MRC from the channel H and the optimal quantized beamforming vector using equation (4). Meanwhile, the transmitter 110 receives at step 360 the index bits from the receiver 130 over the feedback link 150. In step 370, the transmitter 110 decodes the index bits to find the optimal quantized beamforming vector, which the transmitter 110 normalises to unit magnitude. In steps 380 and 390, the transmitter 110 and the receiver 130 apply the normalised beamforming vector and the combining vector ZMRC to transmitted and received data symbols respectively. The channel 125 could be one of the subchannels in a wideband (e.g. Orthogonal

Frequency Division Multiplexing, or OFDM) system, such as 3G LTE and 802.11 a/g WLAN. In such a case, the method 300 could be performed independently for each subchannel of the OFDM channel, or could be performed for only a subset of the OFDM subchannels, and the resulting beamforming and combining vectors could be used to transmit and receive symbols in the respective neighbouring OFDM subchannels.

Design of the Reflected Simplex Codebook

One known codebook uses square lattice angular quantization (SLAQ) to reduce both the search and storage complexity of limited feedback MIMO systems. In SLAQ, the components of codebook vectors are complex integers and can be considered as points of a Quadrature Amplitude Modulation (QAM) constellation. As in QAM, SLAQ codebooks have simple bit-to-symbol indexing algorithms, so codebook storage is not required at either the transmitter or receiver. The problem of searching SLAQ codebooks to estimate the optimal quantized beamforming vector is similar to the problem of noncoherent sequence detection, and fast noncoherent sequence detection algorithms can be utilized to achieve reduction in search complexity orders of magnitude smaller than an exhaustive search. Significantly, the SNR performance of SLAQ codebooks is similar to that of conventional (RVQ) codebooks for the same number of feedback bits, but with much lower computational complexity. Like SLAQ codebooks, Reflected Simplex codebooks adhere to a geometrically regular construction. Reflected Simplex codebook construction is based on a codebook known as Pyramid Vector Quantization (PVQ) that has been used in the context of audio quantization, as part of a very low-delay high-quality speech and audio codec. A Reflected Simplex codebook consists of complex integer vectors mapped from real integer vectors that lie on the original and reflected images about the axes of a (2Nj - l)-dimensional simplex. Reflected Simplex codebooks can be indexed with low time and storage requirements and efficiently searched to estimate the optimal quantized beamforming vector, as described below.

A PVQ codebook D(NJC) consists of real integer vectors of length JV, where K is a parameter that determines the granularity or coarseness of the quantization. The PVQ codebook D(NJC) is constructed from a base set of all real integer vectors that lie on a segment of the hyperplane {Ke_\; ... ; Ke^} that is bounded by the convex hull of the vectors Ke₁ (the vectors e, are the columns of the N by N identity matrix I_N). The hyperplane segment is an (iV-l)-simplex. The parameter K is the height of the simplex, and the length of its edges ΪSKΛFΪ . Note that a 1 -simplex is a line segment, a 2-simplex is an equilateral triangle, and a 3-simplex is a regular tetrahedron. More precisely, the base set is defined by:

In addition, the codebook D(NJQ contains all of the 2 -1 images obtained by reflecting the set (7) about the N axes of R^. The codebook D(NJC) is therefore defined by

Fig. 2 is a plot 200 of the vectors in the PVQ codebook D(NJQ, with N= 2 and AT = 2. The vectors 210 and 220 are the vectors ATe₁ and ATe₂. The hyperplane segment 230 is the 1-simplex. The points (2, 0) 240, (1, 1) 250, and (0, 2) 260 make up the base set (7). The reflections 270, 280, and 290 of the base set {(2, 0), (1, 1), (0, 2)} about the two axes complete the PVQ codebook D(2, 2).

The number of codewords (vectors) in the codebook D(NJC) is denoted as V(NJK). It can be shown that:

= 2N₃F_i(l -K,l -N;2;2)

where ₂Fi is Gauss' hypergeometric function. V(NJK) can also be computed using the following recurrence relation:

F(N₅K) = F(N₅K -I) + F(K₅N -I) + F(N -I₅K -I) (10)

with V(XJK) = 2 for all K and F(N₅I) = 2N. The recurrence relation of equation (10) provides a O(NK) method for computationally efficient calculation of V(NJC) by storing the intermediate values in matrix form, avoiding the multiplications or the complexity of evaluating the hypergeometric function in equation (9). The total number of bits required to transmit a vector in the PVQ codebook is given by CeU(IOg₂(F(N₅K)).

For the case of low-rate quantization, i.e. N» K₅ it can be shown that

1Og₂ F(N₅ K) →

(2N) - iog₂ κ! ₅ (i i)

while for the case of high-rate quantization, i.e. K» N ,

log₂ F(N, K) → (N - l)log₂ (2(K - 1)) + 1 - 1Og₂(N - I)! . (12)

More generally,

iog₂ F(N₅ K) « (N - i)iog₂ (2(κ - I)) - 1Og₂ (N - I)! . (13)

In what follows, the underline notation denotes a bidirectional mapping between a complex vector u of length Nr and a real-valued vector u of length N = 2 Nx defined as follows:

M_2M

= l,...N_τ (14) The PVQ codebook is modified to form the Reflected Simplex codebook as follows. First, construct a modified PVQ codebook C_R(N_J, K) of real integer vectors from the PVQ codebook D(NJC) as follows:

C_R (N_τ,K) = {v 6 D(N,K),v_2r__x > 0,v_2r > 0}, (15) where r = ceil(/ / 2), and / is the index of the first nonzero component of y .

A Reflected Simplex codebook C(Nj, K) is constructed as complex vectors of length NT, derived from the real vectors in C_R(N_J, K) using the mapping in equation (14):

C(N_τ,K) = {v : v ≡ C_R(N_τ,K)} . (16) ^{; !}

The effect of the condition v_2r_, > 0,v_2r > 0in the definition of C_R(N_J, K) (equation (15)) is to remove from C(Nj, K) complex vectors that are phase ambiguous. For example, the effect of the condition on C_R(I, 2) is to leave only the first quadrant vectors in D(2,2), namely (2,0) and (1,1) (see Fig. 2). Two complex vectors V₁, v₂ are phase ambiguous if there exists a complex scalar γ of unit modulus such that V₁= γ V₂. This relation implies that the channel gain (see equation (5)) is the same for both vectors, so one of the vectors is redundant.

If the condition v_2r_, > 0,v_2r > Owere not applied to modify the PVQ codebook, for every v in the Reflected Simplex codebook, there would also be three phase ambiguous vectors -v, j\ and —jv. On removing these redundant vectors, the size of the Reflected Simplex codebook becomes F(N, K) 14, and thus the number of bits required to encode the Reflected Simplex codebook is ceil(log₂(F(N,i£)) - 2. This saving of 2 feedback bits is a significant amount for small codebooks suited to limited feedback beamforming.

Fig. 7 is a flowchart illustrating a method 700 of encoding a real vector of length N to index bits. The method 700 is used at step 330 of the method 300, with its input vector being the real-valued mapping w_opt according to (14) of the optimal beamforming vector W_opt computed at step 320. The method 700 starts at step 710 where the input vector is quantized to the Reflected Simplex codebook C(Nj, K), where N_τ is N I 2, i.e. half the length of the real input vector, and K is the granularity parameter. The step 710, described in more detail below, may alternatively be termed locating the quantized version of an input vector in the Reflected Simplex codebook C(Nj, K). The result is a complex vector from the codebook C(Nj, K). At the next step 720, the method 700 indexes the complex codebook vector to obtain one or more index bits as described in detail below. The method 700 then concludes.

Indexing the Reflected Simplex Codebook Fig. 4 is a flowchart illustrates a method 400 of obtaining an index for a complex vector v in the Reflected Simplex codebook C(Nj, K), carried out by the receiver 130 at step 720 of the method 700 of Fig. 7 when invoked at step 330 of the method 300 of Fig. 3. The indexing method 400 is based on the indexing of the PVQ codebook D(NJC), and has a complexity of 0(NK), mostly avoiding multiplications and divisions (although some operations involve division or multiplication by two, which can be implemented using less complex bit shift operations).

The method 400 begins at decision step 410 where a check is made to determine whether Nj = 0. If so (Yes), the method 400 at step 420 returns an index of zero. If not (No), the method 400 proceeds to step 430, where the components v_t of v are examined in order of ascending t. The value of t is set to the index of the first non-zero component v_t. The value of t indexes a disjoint partition I₍ of the codebook C(Nj, K) in which v lies. The partition I, is in turn made up of K(K+ 1)/2 disjoint subsets l_rJ, where all the vectors v in \_rΛ have the same, unique (non-zero) value of v_t, namely v_r. The subsets l_nt of I, are ordered according to their common v_t values according to the zigzag pattern illustrated in Fig. 5. That is, In contains all the vectors v in I, with v_t =1+0/, while I₂,, contains all the vectors v in I, with v_t =1+1/, and so on.

The size of the subset l_{r t} is given by

At step 440 the index c is set to the sum of the sizes of each partition I₉ for q = l,...t-l, which is given by:

To identify the ordinate / of the subset \_t of I, within which v lies, the method 400 at step 450 sets tf, = «(v, )+3(v, )-l . (19) The method 400 at step 460 then sets / to:

In step 470, the method 400 increments the index c by the sum of the sizes of each subset l_{r t} for r =1 to /-1 using equation (17) for | \_rΛ |. In step 480, the method 400 increments the index c. This is done by recursively invoking the method 400 on the new vector v' that is obtained from v by removing the components v_q for q = 1 to t, also reducing Nj by t and K by Ki, and adding the result to the index c. The method 400 returns the value of c at step 490.

The method for decoding the index bits to a beamforming vector in C(Nj, K), carried out by the transmitter 110 in the step 370 of the method 300 is a straightforward inversion of the indexing method 400. The resulting decoding method also has a complexity of O(NK).

Another form of redundancy arises in the Reflected Simplex codebook between a pair of codebook vectors related by V₁= γv₂ for |/| ≠ 1 , referred to as a divisor ambiguity. Beamforming vectors with divisor ambiguity yield the same effective channel gain (equation (5)) after normalisation to unit magnitude. For example, if N= 2Nj = 4 and K = 4, the vectors [2+/0 2+/0] and [1+yl 1+yl] exhibit divisor ambiguity, with γ = \-j. The simplex-based domain of the PVQ codebook allows calculation of the number of redundant vectors in the Reflected Simplex codebook due to divisor ambiguities. First, divisor ambiguities only occur if K is even. Secondly, iϊ K is even, divisor ambiguity only occurs when y - ±l+j, u_t = ±a+fa; and v, = ±2a+j2a for all t = 1 ,... N. Finally, the number of redundant vectors due to divisor ambiguity is upper bounded by

This formula may be applied to calculate the number of bits required to index the

Reflected Simplex codebook with divisor ambiguities removed. For a large range of N and K, there is no (N, K) pair where removing divisor ambiguities would reduce the integer number of bits. However, for some small K values, a small fraction of a bit is saved, which may provide some efficiencies when a number of vectors are encoded together. A consequence would be more complex index encoding and decoding methods than those described above with reference to Fig. 4, but for small codebooks a lookup table may be used rather than indexing methods.

Searching the Reflected Simplex Codebook The method performed by the receiver 130 in step 710 of the method 700, when invoked at step 330 of the method 300, to locate the optimal quantized beamforming vector in the Reflected Simplex codebook is described hereinafter. A simple exhaustive search over the entire Reflected Simplex codebook can be carried out as in equation (6), but this is prohibitively expensive in terms of computation. Instead, the locating is performed with complexity O(NτlogNj) by observing an equivalence with noncoherent sequence detection.

For the case N_R ≤ 2 , the exhaustive codebook search of equation (6) is equivalent

to finding the closest codebook vector in angle to the optimal unquantized beamforming vector w_opt:

argmax

W =

" opt COS ^{^2} < i9y(iv T,,w TT _oppt_tt ), , (22) v e C{N_T,K) oop

where

When N_R > 2, equation (22) is also a good first-order approximation. Equation (22) is denoted singular vector quantization (SVQ), because equation (22) is equivalent to quantizing w_opt (the right-singular unit vector associated with the largest singular value of the channel matrix H) using an angular metric.

An equivalence relationship exists between SVQ using an angular metric (equation (22)) and the problem of sequence detection over unknown deterministic flat- fading channels. The equivalence can be seen by noting that equation (22) is equivalent to noncoherent detection using the generalized likelihood ratio test (GLRT). Specifically, consider the detection of a complex input vector x, drawn in an i.i.d. manner from a discrete constellation X, given an output vector y:

y = Λx + n , (23)

where n is a vector of i.i.d. white Gaussian noise and h is an unknown complex channel parameter assumed constant over the period of the input vector x. The GLRT-

optimal estimate χ^GLRT of the input vector x is obtained from the received data y by solving

.^ argminmin ₂ _ x e l h ' which becomes, on setting λ = h^'λ,

Algorithms are known for computing χ^GLRT according to equation (25) for any λ

over specific constellations X efficiently, i.e. in polynomial time.

To show the equivalence to equation (22), consider the real-valued mapping w_opt

according to (14) of the optimal beamforming vector w_opt. Denote z as the vector in the

direction of w _t such that the sum of the absolute values of the components of z is K, i.e.

In other words, z is the projection of w_opt onto the reflected simplex that is the

domain of C_R(N₇, K).

Instead of performing the SVQ of equation (22), the following equivalent search

is performed:

argmin ., _||2 w _t = _ϋZ - v , (27)

which in turn is equivalent to the GLRT estimation of equation (25), with

and X= C_R(N₁, K).

Without loss of generality, all the components of z may be converted to be non-

negative. That is, assign

Z₊ = S o Z (28)

where s_n = sgn(z_π )for all n and o denotes the Hadamard product. In other words, Z₊ is the reflection of z onto the simplex on which C_R(N_T, K) is based, i.e. the hyperplane segment defined by

All the codebook vectors in C_R(N_J, K) with non-negative components exist on this simplex. The non-negative portion of the codebook C_R(N_J, K) is a subset of a translated lattice, which is defined as:

In other words, if v e C_R(N,K), thenv₊ - Ke₁ e A_N-1 . The codebook search of equation (27) is therefore equivalent to

w_opt = I- ^™ Wk₊ - Ke₁)-^² + Ke_x . (29)

Equation (29) corresponds to a search for the closest point in Euclidean distance to the vector z₊ - ATe₁ over the lattice A_NA- A method described below with a complexity of O(N log N) for performing the search of equation (29) over the lattice A ._\ is used. The resulting real vector w_opt is mapped using equation (14) to become the optimal quantized beamforming vector w _t .

Fig. 6 is a flowchart illustrating a method 600 of searching the Reflected Simplex codebook to locate the optimal quantized beamforming vector from the optimal beamforming vector w_opt, as used in step 710 of the method 700 of Fig. 7 when invoked at step 330 of the method 300 of Fig. 3. The method 600 begins at step 605, where the real vector w_opt is projected onto the reflected simplex using equation (26) to form a projection z . At step 610, the projection z is reflected onto the non-negative portion of the reflected simplex using equation (28) to form the vector z₊ . At the next step 615, the sum s of the components of Z₊ - KQ₁ is computed, and a new vector z' is computed as:

At the next step 620, the vector z' is rounded to the integer vector f , and the components of f are summed to form the integer variable Δ. At the following step 630, the method 600 computes the difference vector d between z¹ and its rounded version f . A vector t of indices in the range {\,...N} is then formed from d such that t_n is the index of the rt-th most positive component of d.

The method 600 then at step 640 determines whether Δ > 0. If so, step 650 decrements by 1 all the components of f indexed by t_n, for n = 1 to Δ. (These are the indices of the Δ most positive differences between z'and its integer rounded version f .) If not, step 660 increments by 1 all the components of f indexed by t_n, for n = N+Δ+l to N. (These are the indices of the -Δ most negative differences between z¹ and its integer rounded version f .) After both steps 650 and 660, the method 600 returns the complex mapping f of the updated vector f at step 670. An alternative embodiment of step 710 is to perform the method 600 a number (L) of times for different rotations of the vector w_opt . The values of the rotations correspond to phase rotations uniformly spaced between 0 and π / 2. This is achieved by performing the method 600 L times, on w'_opt for / = 0,... Z-I , where:

The result is L candidate beamforming vectors v/'_opi . The candidate beamforming vector that maximises the metric in equation (6) is then chosen for indexing.

In addition to its application in beamforming for MIMO systems, the encoding method 700 may also be applied to any system where a vector needs to be efficiently encoded in a low number of bits. The method 700, for a general input vector, may include the preliminary step of normalizing the vector to be of unit magnitude. The scaling factor is scalar quantized and encoded separately. If the input vector is complex, the mapping of equation (14) should be applied first.

Particular examples include point-to-point precoding such as unitary, linear, and SVD precoding for MIMO systems, where the rows or columns of the channel matrix or precoding matrix need to be sent to the transmitter. Other example systems include multiuser precoding such as zero forcing beamforming and vector perturbation precoding, where the channel vectors for each user need to be sent to the transmitter.

The transmitter 110 and receiver 130 are preferably implemented in dedicated hardware or modules such as embedded integrated circuits performing the functions or sub functions of Figs. 3, 4, 6, and 7, e.g. a PHY/MAC layer chip, or part of a transmitter or receiver chip. The receiver could be a 3 G or 4G cellular handset or enabled PC, or a WLAN enabled handset or PC. Such modules may also comprise graphic processors, digital signal processors, or microprocessors and associated memories.

The arrangements described are applicable to the wireless communication industries.

The foregoing describes only some embodiments of the present invention, and modifications and/or changes can be made thereto without departing from the scope and spirit of the invention, the embodiments being illustrative and not restrictive.

Claims

1. A method of communicating a complex vector, the method comprising: quantizing said complex vector using a codebook, said codebook comprising a plurality of complex vectors mapped from real vectors that lie on a reflected simplex where the simplex is of dimension one less than twice the length of said complex vector; indexing said quantized vector to form one or more index bits; and transmitting said index bits, thereby communicating said complex vector.

2. The method of claim 1, wherein said complex vectors in said codebook have no phase ambiguity.

3. The method of claim 2, wherein said complex vectors in said codebook have no divisor ambiguity.

4. The method of claim 1, wherein said quantizing comprises: mapping said complex vector to a real vector; projecting said real vector onto said reflected simplex; searching over an ANA lattice to estimate a lattice vector that minimises a Euclidean distance from said projected vector; and de-mapping said lattice vector to a complex quantized vector.

5. The method of claim 4, wherein said quantizing further comprises: applying an offset equal to a granularity parameter of said codebook to said projected vector before said searching; and applying said offset to said lattice vector before said de-mapping.

6. The method of claim 4, wherein said quantizing further comprises applying a rotation to said complex vector; repeating said mapping, projecting, searching, and de-mapping on said rotated complex vector to obtain a candidate complex quantized vector; and choosing between said candidate complex quantized vector and said complex quantized vector according to a predetermined metric.

7. The method of claim 1, wherein said indexing comprises: locating a partition of said codebook in which said quantized vector lies; setting an index to the sum of the sizes of partitions of said codebook preceding said located partition; identifying a subset of said located partition in which said quantized vector lies; incrementing said index by the sum of the sizes of subsets of said located partition preceding said identified subset; obtaining a new vector by removing leading zero components from said quantized vector; and incrementing said index by the result of said indexing on said new vector.

8. The method of claim 1, wherein said complex vector is an optimal beamforming vector adapted to be used at the transmitter of a multiple-input wireless communication system.

9. The method of claim 8, further comprising: estimating channel state information for said wireless communication system; and computing said optimal beamforming vector from said channel state information.

10. A method of wireless communication between a transmitter comprising a plurality of transmit antennas and a receiver over a channel, the method comprising: estimating channel state information for said channel; computing an optimal beamforming vector for said plurality of transmit antennas from said channel state information; quantizing said optimal beamforming vector using a codebook, said codebook comprising a plurality of complex vectors mapped from real vectors that lie on a reflected simplex where the simplex is of dimension one less than twice the length of said optimal beamforming vector; indexing said quantized vector to form one or more index bits; and transmitting said index bits to said transmitter.

1 1. The method of claim 10, wherein the receiver comprises a plurality of receive antennas, further comprising: computing a combining vector from said quantized vector and said channel state information, and applying said combining vector to data symbols received from said transmitter on said receive antennas.

12. A receiver adapted to receive wireless communication from a transmitter over a channel, said receiver comprising: an estimation module adapted to estimate channel state information for said channel; a computation module adapted to compute an optimal beamforming vector for said plurality of transmit antennas from said channel state information; a quantization module adapted to quantize said optimal beamforming vector using a codebook, said codebook comprising a plurality of complex vectors mapped from real vectors that lie on a reflected simplex where the simplex is of dimension one less than twice the length of said optimal beamforming vector; an indexing module adapted to index said quantized vector to form one or more index bits; and a transmit module adapted to transmit said index bits to said transmitter.

13. The receiver of claim 12, further comprising: a plurality of receive antennas; a computing module adapted to compute a combining vector from said quantized vector and said channel state information, and a combining module adapted to apply said combining vector to data symbols received from said transmitter on said receive antennas.

14. A system for wireless communication over a channel, the system comprising: a transmitter comprising a plurality of transmit antennas; and a receiver adapted to: estimate channel state information for said channel; compute an optimal beamforming vector for said plurality of transmit antennas from said channel state information; quantize said optimal beamforming vector using a codebook, said codebook comprising a plurality of complex vectors mapped from real vectors that lie on a reflected simplex where the simplex is of dimension one less than twice the length of said optimal beamforming vector; index said quantized vector to form one or more index bits; and transmit said index bits to said transmitter.

15. The system of claim 14, wherein the receiver comprises a plurality of receive antennas and is further adapted to: compute a combining vector from said quantized vector and said channel state information, and apply said combining vector to data symbols received from said transmitter on said receive antennas.