Hochwald construction of unitary matrix codebooks via eigen coordinate transformations

[0001] This patent application claims priority under 35 U.S.C. §119(e) from Provisional Patent Application No. 60/644,375, filed Jan. 14, 2005, the disclosure of which is incorporated by reference herein in its entirety.

[0002] This invention relates generally to wireless communication systems. More particularly embodiments of the invention pertain to the creation and utilization of unitary matrix codebooks.

[0003] Structured limited-feedback closed-loop MIMO preceding methods have been developed for efficiently transmitting data from a mobile station (MS) to a base station (BS) in the context of wireless communication between the two. Such methods do not require the complete feedback of either the preceding weight matrix W or the channel matrix H. Instead, for each pair of transmit antenna number N_t and spatial stream number M_t, a matrix codebook is constructed that is formed of a set of preceding matrices. Both the BS and MS have access to the matrix codebook. Consequently, in order to transmit feedback to the BS, the MS needs only to transmit the index to a preceding matrix contained in the matrix codebook. Such an index is typically formed as a string of bits wherein each bit assumes a binary value. The set of the matrices, or codebook, can be constructed to achieve a desired performance and feedback bandwidth trade-off.

[0004] Once the content of the codebook is fixed, the number of feedback bits that are required to construct the index to the codebook does not grow with the size of the preceding matrix W as is required when implementing existing approaches. It has been shown that methods directed to the structured preceding of codebooks can achieve near-optimal preceding MIMO performance requiring a reasonably low number of feedback bits.

[0005] Central to methods directed to structured MIMO preceding is the design of unitary matrix codebooks. Preferably, such codebooks are designed in such a way that the matrices, or codewords, forming the codebook are as evenly distributed in the Grassmannian manifold as possible. In mathematics, a Grassmannian is the space of all k-dimensional subspaces of an n-dimensional vector space V. One approaches the construction of an optimal codebook as one is better able to maximize the minimum chordal distance of the codebook. As used herein, “chordal distance” refers to the distance between matrices, or codewords, in the multidimensional space comprising the Grassmannian manifold.

[0006] While a codebook generated by the random computation and distribution of codewords within the Grassmannian manifold provides the best possible codebook in terms of minimum Chordal distance, it is not computationally practical to construct such a codebook for systems with large amounts of transmit antennas and spatial streams. To address this constraint, a block-circulant construction method, also called a Hochwald construction method, has been proposed as a systematic construction method which avoids the computational burden of a direct random search method. The Hochwald method provides near-optimal codebooks that facilitate efficient storage and retrieval of the codewords contained therein resulting from the non-computationally intensive parameterization of the codebook. Furthermore, the block-circulant distance structure also facilitates the so-called subspace-tracking method in a precoding based MIMO-OFDM. In the Hochwald design, a codebook is fully specified once the first codeword P_l and a diagonal rotation matrix Q are provided. The other codewords in the codebook are given by (Here 2^L is the cardinality of the codebook and each codeword is a matrix of size N_t×M_t):
P_l=Q^lP_l, for l=2 . . . 2^L,  (1)

[0007] where Q is a diagonal matrix fully parameterized by an integer vector

[0008] $u\stackrel{△}{=}\left[u_1,\dots ,u_{N_t}\right]\text{:}$

[0009] $Q=\left[\begin{array}{ccc}{e}^{j\frac{2\pi }{2^L}{u}_1}& & 0\\ & ⋰& \\ 0& & e^{j\frac{2\pi }{2^L}{u}_{N_t}}\end{array}\right]$

[0010] Furthermore, in this design, the first codeword P_l is chosen to be a N_t×M_t sub-matrix of the N_t×N_t Discrete Fourier Transform (DFT) matrix D_Nt whose (m,n) element is specified as

[0011] ${\left(D_{N_1}\right)}_{m,n}=e^{j\frac{2\pi }{N_t}\left(m-1\right)\left(n-1\right)}$
where 1≦m,n≦N_t. Denoting d_c as the c^th column of the matrix D_Nt, the first codeword is the collection of M_t columns parameterized by the set of column indices

[0012] $c\stackrel{\Delta }{=}\left[c_1,\dots ,c_{M_t}\right],i.e,P_1=\left[d_{c_1},\dots ,d_{c_{N_t}}\right].$
With reference to Table 1 below, there is tabulated in exemplary fashion the choices of

[0013] $u\stackrel{\Delta }{=}\left[u_1,\dots ,u_{N_t}\right]\mathrm{and}c\stackrel{\Delta }{=}\left[c_1,\dots ,c_{M_t}\right]$
for different transmitter antenna numbers N_t between two and four, and spatially multiplexed data stream numbers M_t, subject to the constraint that N_t>M_t. Note that the choice of 2^L, the codebook size, is the result of a trade of between performance and the number of required feedback bits.

[0014]

TABLE 1
Nt (# of Tx Antennas)	Mt (# of data streams)	2L/(L) Codebook size/ (feedback bits)	c ⁢ = Δ [ c 1 , … ⁢ , c M t ] Column indices	u ⁢ = Δ [ u 1 , … ⁢ , u N t ] Rotation Vector
2	1	8/(3)	[1]	[1, 0]
3	1	32/(5)	[1]	[1, 26, 28]
3	2	32/(5)	[1, 2]	[1, 26, 28]
4	1	64/(6)	[1]	[1, 8, 61, 45]
4	2	64/(6)	[0, 1]	[1, 7, 52, 56]
4	3	64/(6)	[0, 2, 3]	[1, 8, 61, 45]

[0015] Note that each column index 0<=c<=Nt−1, c=[0] indicates the first column of the DFT matrix D, c=[1] indicates the second column of the DFT matrix D, etc.

[0016] Unfortunately, codebooks formed of codewords that are generated in accordance with the Hochwald construction method are subject to various undesirable constraints. Amongst these constraints are the requirement that the energy of any two elements within a single matrix codeword be equal and the total energy of each individual codeword be equal to the total energy of every other individual codeword. While such constraints allow for a straightforward application of the Hochwald construction method, their enforcement results in a distribution of the codewords of a codebook in the Grassmannian manifold that can be improved upon.

[0017] It would therefore be preferable to derive a method for performing Hochwald construction that does not suffer from the constraints of the prior art.

[0018] In accordance with an example of the present invention, a method of deriving a codebook comprises deriving a first codeword P_l in Euclidean coordinates, transforming the first codeword into eigen-coordinates, applying a Hochwald construction to the first codeword in eigen-coordinates to derive a plurality of codewords in eigen-coordinates, and transforming the plurality of codewords in eigen-coordinates into a plurality of codewords in Euclidean coordinates to form the codebook.

[0019] In accordance with another example of the invention, a data structure embodied on a tangible medium comprises a first codeword P_l, and a plurality of codewords P_l, l=2 . . . 2^L, where 2^L comprises a number of the plurality of codewords wherein P_l=VQ^lV^HP_l where V comprises a unitary matrix comprising eigen vectors, Q comprises a diagonal matrix, and V^H comprises a hermitian eigen matrix.

[0020] In accordance with another example of the invention, a mobile station comprises a means for storing a codebook the codebook comprising a first codeword P_l, and a plurality of codewords P_l, l=2 . . . 2^L, where 2^L comprises a number of the plurality of codewords wherein P_l=VQ^lV^HP_l where V comprises a unitary matrix comprising eigen vectors, Q comprises a diagonal matrix, and V^H comprises a hermitian eigen matrix, and a means for accessing the first codeword and the plurality of codewords.

[0021] In accordance with another example of the invention, a base station comprises a means for storing a codebook the codebook comprising a first codeword P_l, and a plurality of codewords P_l, l=2 . . . 2^L, where 2^L comprises a number of the plurality of codewords wherein P_l=VQ^lV^HP_l where V comprises a unitary matrix comprising eigen vectors, Q comprises a diagonal matrix, and V^H comprises a hermitian eigen matrix, and a means for accessing the first codeword and the plurality of codewords.

[0022] In accordance with another example of the invention, a program of machine-readable instructions, tangibly embodied on an information bearing medium and executable by a digital data processor, to perform actions directed toward the transmission of information, the actions comprising obtaining an information matrix, accessing a codebook to select one of a plurality of indexes corresponding to the information matrix the codebook comprising a first codeword P_l, a plurality of codewords P_l, l=2 . . . 2^L, where 2^L comprises a number of the plurality of codewords wherein P_l=VQ^lV^HP_l where V comprises a unitary matrix comprising eigen vectors, Q comprises a diagonal matrix, and V^H comprises a hermitian eigen matrix wherein each of the first codeword and the plurality of codewords corresponds to a unique one of the plurality of indexes, and transmitting the one of the plurality of indexes.

[0023] In accordance with another example of the invention, a program of machine-readable instructions, tangibly embodied on an information bearing medium and executable by a digital data processor, to perform actions directed toward the reception of information, the actions comprising receiving an index, and accessing a codebook to select one of a plurality of codewords corresponding to the index the codebook comprising a first codeword P_l, a plurality of codewords P_l, l=2 . . . 2^L, where 2^L comprises a number of the plurality of codewords wherein P_l=VQ^lV^HP_l where V comprises a unitary matrix comprising eigen vectors, Q comprises a diagonal matrix, and V^H comprises a hermitian eigen matrix wherein each of the first codeword and the plurality of codewords corresponds to a unique one of the plurality of indexes.

[0028] In an example of the present invention, there is provided an eigen-coordinate transformation based method to improve the minimum Chordal distance of the codebooks generated by the Hochwald construction method. This is achieved, in part, by relaxing at least one constraint of the Hochwald construction so as to render more nearly optimal the distribution of codewords in the multi-dimensional Grassmannian manifold. As noted previously, utilization of the Hochwald construction method results in a codebook wherein the energy distribution among different elements in each individual matrix codeword is equal and fixed from one matrix codeword to every other. That is, for any two matrix codewords P_l1 and P_l2, the (i,j) element of each codeword has the same energy, i.e. ||(P_l1)_i,j|²=|(P_l2)_i,j|².

[0029] In an example of the present invention, this constraint is removed to provide a method for generating codebooks with better Chordal distance properties. Specifically, the constraint is relaxed so as to obtain codebooks wherein the minimum Chordal distance of the codebook is more nearly maximized in comparison to instances wherein the constraint is not relaxed. The eigen-coordinate transformation method recited herein removes the above noted constraint by taking advantage of the energy-shifting property of eigen-coordinate transformations while preserving advantageous features of the Hochwald construction such as block-circulant distance distribution of codewords and easy parameterization.

[0030] In the following description of an example of the invention, reference is made to the accompanying drawings which form a part hereof, and in which is shown by way of illustration of an exemplary fashion in which the invention may be practiced. It is to be understood that other examples may be utilized, as structural and operational changes may be made without departing from the scope of the present invention.

[0031] As noted above, an example of the present invention discloses a method utilizing an eigen-coordinate transformation to provide a systematic way of removing the fixed-energy-distribution constraint in the Hochwald construction. Such a method more nearly maximizes the minimum Chordal distance of a codebook constructed therefrom.

[0032] Upon examination of equation (1), it is evident that all of the other codewords P_l=Q^lP_l, for l=2, . . . 2^L are derived from the first codeword P_l. This observation makes evident the utility to be gained by removing the fixed-energy distribution constraint by allowing P_l to be an arbitrary unitary matrix, rather than the columns of a DFT matrix as defined in the Hochwald Construction.

[0033] However, removal of such a constraint does not fully resolve the shortcomings of the Hochwald Construction as the energy distribution is still fixed from one codeword to another. This is true even though the energy is no longer uniformly distributed within one codeword, as is the case in the Hochwald construction. Therefore, changing the first codeword P_l to be an arbitrary unitary matrix alone is not sufficient to most advantageously optimize the construction of a codebook.

[0034] Therefore, an example of the present invention utilizes an eigen-coordinate transformation to achieve variations in energy distribution among different matrix codewords. First, since Q is a diagonal matrix, it can be represented as a part of an eigen-decomposition of some matrix S:

[0035] $\begin{array}{cc}S\stackrel{\Delta }{=}{\mathrm{VQV}}^H& \left(2\right)\end{array}$
where V is a unitary matrix of size N_t×N_t consisting of the resultant eigen vectors. Next, the matrix Q can be replaced with matrix S in equation (1) to arrive at an improved Hochwald construction:
P_l=S^lP_l=(VQV^H)^lP_l=VQ^lV^HP_l, for l=2, . . . 2^L,  (3)
where the parameterization of eigen-matrix V is discussed more fully below.

[0036] With reference to FIG. 1, there is illustrated a flow chart of an example of the improved Hochwald construction of the present invention. First, the first codeword, p_l, is transformed into the first codeword in eigen-coordinates at block 11 by applying the hermitian eigen-matrix V^H to p_l to yield V^HP_l. The Hochwald construction is then applied to the first codeword in eigen-coordinates, V^HP_l, at block 13 to derive the lth codeword in the eigen-coordinate, Q^lV^HP_l. The lth codeword in the eigen-coordinate is then transformed back to the Euclidean coordinate with matrix V at block 15 to generate each final codeword P_l=VQ^lV^HP_l.

[0037] Because this methodology applies the Hochwald Construction to the first codeword transformed into eigen-coordinates, the intermediate codewords Q^lV^HP_l, for l=1, . . . , L still have the same energy distribution in the eigen-coordinates defined by V. However, the final codewords, P_l=VQ^lV^HP_l, will have different energy distributions once transformed into Euclidean coordinates. For vector codebooks, it is sometimes convenient to remove the phase of the first element in each codebook by {tilde over (P)}_l=P_le^−jφi for l=2, . . . , 2^L, where φ_i is the phase of the first entry of P_l. In this alternative representation of codebook that is strictly equivalent to the original codebook, the first element of each codeword is real.

[0038] With reference to FIG. 2, there is illustrated in exemplary fashion, the different energy distributions amongst elements of a codeword produced by the energy-shifting properties of eigen-coordinate transformation. Points A and B are elements within a single codeword derived according to an example of the present invention. The x and y axes represent Euclidean space while eigen vectors v1 and v2 represent eigen space. The coordinates of point A are ((sqrt(3)/2), ½) in eigen space, and (0, 1) in Euclidean space. Similarly, the coordinates of point B are (−(sqrt(3)/2), ½) in eigen space, and (−(sqrt(3)/2), −½) in Euclidean space. As a result, the energy distribution of points A and B are seen to be the same in eigen-coordinate space defined by v1, and v2, but different in Euclidean coordinate space.

[0039] It remains to be shown how the eigen-matrix V is parameterized. Here we further impose the Householder structure on V to allow a very simple parameterization of V, i.e.

[0040] $\begin{array}{cc}V\left(a\right)\stackrel{\Delta }{=}\{\begin{array}{cc}I,& a=e_1\\ I-\mathrm{pw}•w^H& \mathrm{otherwise}\end{array}& \left(4\right)\end{array}$

[0041] where w=a−e₁ and e₁[1 0 . . . 0]^T and

[0042] $p=\frac{2}{w^{H}w}.$
I is an identity matrix and the superscript H denotes Hermitian Conjugate. Given this additional parameterization of vector a, the Codebook is now uniquely determined once the three vectors u, c and a are obtained. We denote this codebook as P(u,c,a). The search for an optimal codebook parameterized by these three vectors is described by:

[0043] $P^{\mathrm{optimal}}\left(u,c,a\right)=\underset{u,c,a}{\max }\left\{\underset{P_i,P_j\in P\left(u,c,a\right)}{\min }{d}_{\mathrm{chordal}}\left(P_i,P_j\right)\right\}$

[0044] Where d_chordal(P_i,P_j) denotes the Chordal distance between two codewords P_i,P_j. The optimal codebooks for various antenna configurations are tabulated in Table 2. With reference to Table 2 below, there is illustrated the effectiveness of the method of the present invention showing the parameters of several codebooks using the improved Hochwald construction described herein. Note that the first codeword P_l remains formed as columns of the DFT, as is the case in an unadulterated application of the Hochwald construction. The eigen vector a has been added as an additional parameter. In an example of the present invention there is practiced joint optimization over vector a and the first codeword P_l, which is, as noted above, an arbitrary unitary matrix. Of particular interest is the last column wherein there is stated the maximum minimum Chordal distance between the codewords of the codebook using the improved Hochwald construction of the present invention versus that of the standard Hochwald contruction. In all instances, the minimum Chordal distance is improved through the implementation of the method of the present invention.

[0045]

TABLE 2
Nt	Mt	2L/(L)	c ⁢ = Δ [ c 1 , … ⁢ , c M t ] Column indices	u ⁢ = Δ [ u 1 , … ⁢ , u N t ] Rotation Vector	Vector a	Chordal distance (Improved Hochwald)/(Hochwald)
2	1	8/(3)	[1]	[1, 4]	[1 0]	0.1464/0.1464
3	1	8/(3)	[1]	[1, 4, 3]	[1 0 0]	0.6667/0.6667
3	1	64/(6)	[1]	[1 26 57]	[1.2518 − 0.6409i	0.1263/0.1166
					−0.4570 − 0.4974i
					0.1177 + 0.2360i]
4	1	8/(3)	[1]	[1, 2, 7, 6]	[1.2895 + 0.3635i	0.8282/0.7500
					0.5287 − 0.2752i
					−0.2352 − 0.4247i
					−0.4040 + 0.1729i]
4	1	64/(6)	[1]	[1 45 22 49]	[1.3954 − 0.0738i	0.3935/0.3643
					0.0206 + 0.4326i
					−0.1658 − 0.5445i
					0.5487 − 0.1599i]
5	1	64/(6)	[1]	[1 17 40 6 39]	[1.1052 − 0.1146i	0.5864/0.5642
					−0.5410 − 0.3587i
					−0.0773 + 0.6108i
					−0.1388 − 0.0375i
					0.3007 + 0.2538i]
6	1	64/(6)	[1]	[1 24 60 46 49 39]	[0.9859 + 0.6880i	0.6810/0.6609
					0.3354 − 0.2254i
					0.2202 + 0.1981i
					−0.0912 + 0.2780i
					0.3314 + 0.1186i
					0.1675 − 0.1946i]

[0046] Note that each column index 0<=c<=Nt−1, c[0] indicates the first column of the DFT matrix D, c=[1] indicates the second column of the DFT matrix D, etc.

[0047] Note that, in Table 2, vector a is precisely defined as being formed of both real and imaginary parts. The present application is drawn to all implementations of vector a, as well as deviations from the values of vector a, whether intentional or inadvertent, which might arise in operation such that an implemented vector a is sufficiently close to the value recited above so as to enable use of the codebook.

[0048] Once computed, a codebook constructed according to an example of the present invention can be utilized as illustrated with reference to FIG. 3. Base station 35 is in wireless communication with mobile station 33. Mobile station 33 is any mobile device capable of wireless communication including, but not limited to, wireless telephones, personal digital assistants (PDAs), personal computing devices, and the like. Both base station 35 and mobile station 33 are formed of a digital signal processor (DSP) 37 or another type of data processor for accessing and operating upon digital data. DSP 37 is coupled to a memory device 31 capable of storing digital data. A codebook of the present invention is preferably stored upon each memory device 31 coupled singularly to the base station 35 and the mobile station 33.

[0049] The DSP 37 of mobile station 33 selects the index for the codeword corresponding to the information that the mobile station 33 wishes to transmit to base station 35. Examples of the kind of data corresponding to the codewords of the codebook include, but are not limited to, channel information in the form of channel matrices, data receipt acknowledgement, precoding weight matrices, and SNR measurements on spatial directions. Preferably, the index is a string of data comprised of a number of binary bits b. As each codeword corresponds to one and only one index, the maximum number of codewords in a codebook is equal to 2^b.

[0050] Typically, there will not exist a codeword that is precisely equal to the information to be transmitted. Typically, such information is in the form of a matrix so as to form an information matrix. It is for this reason that the methodology of the present invention serves to evenly distribute, or, more precisely, to maximize the minimum chordal distance between, codewords. By so doing, information can be transmitted by sending the index of the codeword that is closest to the information matrix, such as for example a channel matrix, in the multidimensional Grassmannian manifold with a minimum loss of accuracy. The DSP 37 of the mobile station functions to perform a calculation or calculations as directed by software or hardware encoded logic to select the codeword of the codebook that is closest to the information to be transmitted so as to obtain the index corresponding to the codeword so selected.

[0051] The index so derived is typically transmitted by the mobile station 33 by at least one transmit antenna 41 via a feedback loop 39 to the base station 35. The base station 35 receives the transmitted index through the use of at least one receive antenna 43. As noted, base station 35 is similarly equipped with a DSP 37 in communication with a memory device 31 upon which is stored the codebook 45. Using the received index, the DSP 37 of the base station operates through the execution of software or hardware encoded logic to retrieve the corresponding codeword from the codebook 45.

[0052] As noted above, the examples of this invention may be implemented by computer software executable by a data processor of the MS 33, or by hardware circuitry, or by a combination of software and hardware circuitry. The MS may be one component of a wireless communication system. Further in this regard it should be noted that the various blocks of the logic flow diagram of FIG. 1 may represent program steps, or interconnected logic circuits, blocks and functions, or a combination of program steps and logic circuits, blocks and functions for performing the specified tasks.

[0053] With reference to FIG. 4, there is illustrated a flow chart of an example of the method of the invention.

[0054] Although described in the context of particular examples, it will be apparent to those skilled in the art that a number of modifications and various changes to these teachings may occur. Thus, while the invention has been particularly shown and described with respect to one or more preferred embodiments thereof, it will be understood by those skilled in the art that certain modifications or changes, in form and shape, may be made therein without departing from the scope and spirit of the invention as set forth above.