GB2475098A - Transmission of channel state information (CSI) with reduced overhead - Google Patents

Abstract

A process for compressing channel state information (CSI) to be fed back to a transmitter comprises arranging data as a matrix comprising a number of orthonormal vectors derived from a channel matrix, determining a singular value decomposition (SVD) of a subset of the orthonormal matrix to generate matrices respectively of left and right singular vectors, the number of vectors in the subset being equal to the order of the vectors, and right multiplying the remainder orthonormal vectors not included in the singular value decomposition by a matrix product of the matrix of right singular vectors and the matrix of left singular vectors to generate a matrix of compressed data. A decompression technique is also disclosed. The process allows for the reduction of the number of coefficients that are required to be encoded and is achieved with little added complexity (one small SVD operation).

Description

WIRELESS COMMUNICATION APPARATUS AND METHOD

The present invention is concerned with signalling channel state information in a wireless communication network. It is particularly, but not exclusively, concerned with compression of channel state information to reduce overhead required for transmission of the same.

Wireless communications standards are in a continuous cycle of revision and development, as improvements to various communications techniques become known.

One strand of telecommunications standards to which this statement applies is the 3GPP series of standards. Revision 10 of this series is known as LTE-Advanced (Long term evolution).

Spectral efficiency is an area of concern in this field of technology. This is a measure of the extent to which an available bandwidth is being used to its theoretical maximum information capacity. Naturally, no practical technique reaches this maximum, but it is desirable that the efficiency be as high as possible, with due regard to local constraints such as computational complexity, power consumption and so on. This is becoming particularly relevant in view of the considerable growth in usage of wireless communication, which has increased demand for bandwidth and therefore has made higher spectral efficiency more desirable.

Many wireless communication techniques require the communication of information to establish and maintain a communication channel, or to aid in reception of a signal.

That is, in order to establish a viable communication channel, to communicate a piece of information, other information needs to be communicated as well. This inevitably reduces channel efficiency to a level below ideal. It is thus desirable to minimise, to the extent this is possible, the amount of non-content information to be communicated across the communications channel.

Downlink multi-antenna transmission techniques have been identified as key enablers of high spectral efficiency requirements in the 3GPP LIE-Advanced standard, with peak throughput of up to 30 bit/s/Hz. These techniques include enhanced single-cell single-user (SU) and multi-user (MU) MIMO supporting multiple spatial layers, and coordinated multipoint (C0MP) transmission.

CoMP involves the cooperation of two or more base nodes in scheduling resources and/or generating and broadcasting a signal for downlink transmission. Two main C0MP modes have been identified, namely cooperative beamforming (GB) and joint processing (JP).

In GB, a cooperating cell that is equipped with multiple transmit antennas chooses transmit beam(s) by suitably trading between precoding gain to a served terminal (user equipment -UE) and interference reduction with respect to a victim' UE served by an adjacent cell. Similarly, JP involves serving data packets to multiple UEs from multiple cell sites, wherein antenna weights at different cells are chosen to achieve simultaneous coherent channel combining and transmit interference nulling.

Deployment scenarios in which CoMP is deemed to deliver a substantial boost in network capacity and user experience include high interference heterogeneous networks, where the traditional macrocell network coexists with one or more types of non-traditional nodes such as femto-cells in closed subscriber loop (CSG), hotzone (pico) cells in open subscriber group (OSG) and relay nodes lacking a wired backhaul connection.

Similar to closed loop precoding in LIE (Release 8), transmit interference nulling relies on channel state feedback from the UE. However, interference nulling requires more accurate channel information in comparison with basic LIE Rel 8 precoding, as widely acknowledged by many recent contributions in the 3GPP RAN1, for example Ri- 094217, "Feedback in support of downlink CoMP: general views," Qualcomm, 3GPP TSG RAN 1#S8bis, Oct 2009.

The arrangement defined by LIE Release 8 does not provide for an explicit channel feedback report. Instead, in that case, the channel feedback report is implicit in that the UE tests certain hypotheses on the transmission and/or reception spatial processing and then signals a codebook entry corresponding to the hypothesis that yields the highest received SNR (3GPP TS 36.211 v8.8.0, § 6.34). This form of channel information is effective in maximising the beamforming gain (as in SU-MIMO) but is very restrictive for interference nulling, because a UE cannot predict the interference experienced by other UEs. In fact, one aspect of this area of technology which is currently under discussion is that of providing, on an efficient basis, explicit information on single-cell and multi-cell spatial channel structures as well as channel quality information needed to facilitate scheduling and link adaptation by the network.

Explicit feedback schemes are currently being explored for their suitability for inclusion in a further revision of the LTE-Advanced standard. One possibility is to perform either a joint or a separate quantisation of the strongest eigenvectors and eigenvalues of the Gram matrix of the channel, Hilt, where H is the nxm channel matrix form receive antennas and n transmit antennas and the symbol t denotes Hermitian transposition.

Alternatively, joint or separate quantisation has been proposed for the strongest right singular vectors and singular values of H such as in R1-094041, "SU/MU MIMO feedback with codebook-based vector quantization," Samsung, 3GPP TSG RAN 1#58bis, Oct 2009.

One shortcoming of explicit feedback is that it leads to an increased overhead to convey spatial information to the transmitter. This may outweigh the benefit provided by throughput increase made possible by interference nulling. In fact, in order to provide sufficiently accurate information, multiple channel state reports are needed across the allocated bandwidth and in time; for each report it may be desirable to signal multiple eigenvectors. Besides, it may also be desirable for the UE to report some implicit feedback to support MIMO schemes in situations where maximising the beamforming gain is more important than interference nulting, such as in single-cell SU-MIMO. Finally, some channel quality indicators (CQI) are also needed for link adaptation and possibly user selection. Therefore, it is desirable to adopt a compression mechanism in order to reduce the overhead of reporting the channel explicit spatial structure.

An aspect of the present invention provides a method of compressing data for transmission across a communications channel, the data representing channel conditions for the communications channel in the opposite direction, comprising arranging the data as a matrix comprising a number of orthonormal vectors derived from a channel matrix, determining a singular value decomposition of a subset of the orthonormal matrix to generate matrices respectively of left and right singular vectors, the number of vectors in the subset being equal to the order of the vectors, and right multiplying the remainder orthonormal vectors not included in the singular value decomposition by a matrix product of the matrix of right singular vectors and the matrix of left singular vectors to generate a matrix of compressed data.

Another aspect of the invention provides a method of decompressing compressed data, said data being arranged as a square matrix, comprising determining a singular value decomposition of the square matrix, constructing a matrix Y' comprising: 1/2 ,, v(I_) Vt U1zy1t where V1 is a right singular value matrix resultant from said singular value decomposition, U1 is a left singular value matrix resultant from said singular value decomposition, and is a diagonal matrix resultant from said singular value decomposition.

An aspect of the invention provides a method for compressing the representation of a set of channel eigenvectors, in a channel state information (CSI) feedback scheme, whereby the original number of coefficients is reduced without distorting the vector space spanned by the set of eigenvectors.

The method may comprise jointly encoding a set of p n-dimensional eigenvectors, with n »= p, by using n-p coefficients per vector.

The encoding may comprise arranging the set of eigenvectors in an n x p orthonormal matrix, taking a singular value decomposition (SVD) of the top p x p block of the matrix, and multiplying the bottom (n-p) x p matrix block by the SVD output to determine a channel representation.

Another aspect of the invention comprises a method of decoding a channel representation comprising reversing the above operations to reconstruct the orthonormal n x p matrix from a (n-p) x p block.

Although all of the above examples specify apparatus supplied to perform specific operations, the reader will appreciate that other aspects of the invention may comprise computer program products. For instance, a computer program product could be provided to configure a computer to operate as a transmitter as previously described.

A computer program product could be provided to configure a general purpose radio communications device to operate as a receiver as previously specified. This particularly applies to a general purpose radio communications device suitable for configuration in line with the general provisions of software defined radio.

The computer program product could be provided as a storage means, storing computer executable instructions. Alternatively the computer program produce could be provided as a signal receivable by a computer to cause the computer to become suitably configured.

The computer program product may comprise instructions representative of a complete computer program, or may comprise instructions representative of an update to an existing computer program.

Specific embodiments of the invention will now be described, with reference to the accompanying drawings, in which: Figure 1 is a schematic diagram of a wireless communications network in accordance with a first embodiment of the invention; Figure 2 is a schematic diagram of a wireless communications device in accordance with the first embodiment of the invention; Figure 3 is a flow diagram of a process for generating and using channel state feedback information in accordance with the first embodiment of the invention; and Figure 4 is a flow diagram of a process of matrix reduction in the process illustrated in figure 3; and Figure 5 is a flow diagram of a process of matrix reconstruction in the process illustrated in Figure 3.

In general terms, the specific embodiment of the invention is illustrated in Figures 1 and 2. Figure 1 illustrates a very schematic wireless communications system 10 comprising first and second transmit/receive devices 20, 30. Each of the transmit/receive devices 20, 30 is a multi antenna device, and antenna selection is well known to be a suitable way of making best use of the channel which can be formed between two such devices.

In use, of course, it will be understood that one or other of these transmit/receive devices may be established as a base station in a cellular network but the exact configuration of a network is not a critical aspect of the present embodiment of the invention.

Further illustrating this arrangement, Figure 2 is a schematic diagram of the transmit/receive device 20 illustrated in Figure 1. The device 20 comprises a transmitter driver 22 operable to receive data from a data source/sink 26 for transmission on the antennas TX1 to TX4 of the device 20. A receiver driver 24 receives and processes signals received on the same antennas and passes data to the data source sink 26 and channel state information to the transmitter driver 22. An array of suitable switches is provided to enable the antennas to be used for both transmitting and receiving.

The transmitter driver 22 is configured to transmit, as frequently as appropriate to the application, channel state information, to be received (fed back) by a corresponding device and processed by the corresponding receiver driver of that device. The structure of that channel state information, in accordance with the specific embodiment of the invention, will now be described.

nxl eigenvectors y1,...,y., are defined, representing the channel spatial structure to be fed back, and Y is defined as the ii x p orthonormal matrix whose columns are the p eigenvectors. These can be obtained, for example, as the p strongest eigenvectors of the matrix product Hilt, or some average E(HHt),where H is the baseband representation of a flat fading downlink channel between a base station node (eNodeB) and a UE. Alternatively, the p vectors can be associated with the strongest right singular vectors of H, or some average E(H).

The p vectors are orthogonal and unit-norm. If a Standard mandates that they should be represented separately, without considering their orthogonality, the condition on the norm imposes one constraint per vector. Therefore, the degrees of freedom associated with each vector representation, i.e. the number of (complex) coefficients required to represent each vector is n-i. It will be appreciated that, without loss of generality, it can be assumed that the first element of each eigenvector (or singular vector) is real-valued. Collectively, np-p (1) coefficients must be encoded to represent the p vectors.

The above representation is redundant as the orthogonality constraints between the vectors is not considered. In fact, if a Standard mandates that the vectors should be represented jointly, for the first vector the norm still imposes a single constraint, for the second vector one constraint is imposed by the norm and one by the orthogonality to the first vector, for the third vector there are two orthogonality constraints and one norm constraint and so forth for the remaining vectors. Therefore, the overall degrees of freedom in the representation are: (n -i) -p(p + . (2) This is a more efficient way of conveying the same information as in the separate representation.

In the present embodiment, a yet further efficient representation of the subspace spanned by the p vectors is used. This is based on the observation that, for the purpose of interference nulling, the transmitter does not need exact knowledge of the orthonormal matrix Y. Instead, the vector sub-space spanned by its columns is sufficient. In other words, the transmitter should be able to reconstruct an arbitrary linear combination of the columns of Y, i.e. a matrix Y' YQ, where Q is an arbitrary p x p unitary matrix, unknown to the transmitter.

The reader will appreciate the reason why this does not affect the nulling capability of the transmitter. A precoding vector belongs to the null-space of V if and only if it belongs to the null space of Y'* Therefore, as it is desirable for the transmitter to form a beam that does not interfere with the channel space represented by the eigenvectors of Y, it can equivalently choose a vector from the nufl space of Y or Y'* This MIMO transmit processing requirement is common to many configurations, such as single-cell MU-MIMO with zero-forcing precoding or multi-cell (CoMP) cooperative beamforming.

The interference nulling capability may also be the preferred interference reduction technique in heterogeneous deployments of macro-cells and femto-or hotzone-cells.

The degrees of freedom of this sub-space representation are obtained by subtracting from the degrees of freedom of Y (given by equation (2)), those of the square Q matrix, also given by an equation of the form of equation (2) but replacing n with p).

Thus, the number of (complex) coefficients associated with the new representation is (np_P(I+1)J_(p2 _P(P+1)J__np_p2 (3) It is self evident that (np-p2) <(np-P(P+I)) <(np-p) forp>1.

More specifically, if a unit cost is associated with the representation of each coefficient, for example number of feedback bits per coefficient, and the initial vector-by-vector representation is established as a baseline, the percentage overhead reduction achievable with the subspace representation of the present embodiment can be determined. This is set out in Table 1, which sets out the degrees of freedom and overhead reduction in the representation of the spatial structure of a MIMO channel. In the table, n is the number of transmit antennas, and p the number of reported channel eigenvectors. The sub-space representation of the present specific embodiment is compared to separate and joint eigenvector representation (prior art):

Table 1.

Separate eigenvector Joint Sub-space representation(baseline) eigenvector representation representation (this invention) Degreesof n-i p+i n-.

freedom per 2 eigenvector (when reporting p eigenvectors) Overhead 0 p-i p-i reduction with 2(n-1) n-i respect to baseline It is notable that, with the arrangement as set out in the above described specific embodiment, the equivalent overhead of (p-i) coefficients can be saved, as opposed to previous arrangements, for each eigenvector representation, when signalling p eigenvectors. (p.1)12 coefficients are saved per vector on the more efficient joint eigenvector representation.

Figure 3 describes in general terms the stages of data processing and transmission which are undertaken in order to furnish a transmitter with channel state information in accordance with the specific embodiment of the invention.

At the receiver there is an initial step 32 of gathering channel state information on the signal received at the receiver. This may be as an instantaneous measurement, or an average over a period of time.

Then, singular value decomposition is performed (34) on the resultant channel matrix.

The p strongest right singular vectors are chosen for further processing, designated as the matrix Y for further processing.

From a structural perspective, the distinction between the feedback generation mechanism of the present embodiment, and that of the prior art, is in a compression stage 36 carried out before a quantisation block 38. This operation is named "matrix reduction" in the block diagram of figure 3 and will now be described.

This matrix reduction operation adds little complexity to the feedback generation procedure at the UE as will be understood from the following description thereof; it comprises mainly an additional singular value decomposition (SVD) performed on a small pxp matrix.

The reverse operation required at the transmitter to expand the encoded feedback, after reconstruction, into a set of orthonormal vectors, is depicted in Figure 3 as a "matrix expansion" block. This block also entails an SVD operation on an (n_p)xp matrix, which amounts to a small increase in the transmitter complexity.

The "matrix reduction" operation (36) of Figure 3 will now be described in further detail.

It should be noted that the number of reported eigenvectors, p, cannot be greater than the number of transmit antennas, and therefore it can be assumed that The matrix V containing the set of p orthonormal vectors is input to the block (step sl-2 in figure 4) and is partitioned as follows where is a P>< P matrix consisting of the first p rows of V, while contains the V. remaining n-p rows. The SVD (singular value decomposition) of " is then taken (step S1-4): = (4) The new reduced-size (n -p) x p matrix to be quantised and fed back is given by I' = X_PVCV (step S1-6) where (by definition) VCVt =Q is a pxp unitary matrix.

The resultant matrix F is then output (step S1-8) to the source coding and quantization stage (38).

At the transmitter side, the received feedback is reconstructed (42) and then the reconstructed feedback data matrix is re-expanded (44). The resultant matrix of orthonormal vectors can then be used in determining a precoder (46) for further MIMO transmissions.

The "matrix expansion" operation (44) at the transmitter side of the feedback link is carried out as follows. For notational convenience it is assumed that the quantisation and reconstruction blocks do not introduce any distortion on 1', which is input to the process in step S2-2 in figure 5. The matrix expansion block takes the compact SVD of the reconstructed I' (step 82-4): 1'-_U1V1 (5) and computes the orthonormal nxp matrix (step S2-10), if n »= 2p, as follows: , v(i2) \) (6) U1cVjt which is then output in step S2-12.

The key property is that the columns of Y' and Y span the same subspace, namely it can be shown that: Y'=YQ (7) To show this, it should be observed that, in equation (4), the singular values are the cosines of the principal angles between the sub-space spanned by the columns of (I " V and the p reference axes given by the columns of the matrix. By taking °(n-p)xp) the compact SVD of the block V,2...,, = uv7, is obtained, where the singular values are the sines of the principal angles identified above and V is obtained from a permutation and possible sign change of the columns of V. Thus, vs = vcP, with P generalised permutation matrix, such that P = P". It should be noted that, by definition of the sine and cosine function, P(I _)2p' = By plugging the above SVD into the definition of F, the result is: = JpHvH and by comparison with (5), it follows that V1 = VPD and U1 = UD, where D is a diagonal matrix with complex exponentials on the diagonal, such that W' = Dh'.

Finally, the top block in (6) can be rewritten as follows: V1 (I _)hI2 V1? = VPD(I -4)U2 D1P'V' VV YQ, which, along with the definition of F, proves (7).

By using this procedure, the spatial information embedded in Y can be conveyed with the fewest coefficients: p(n P), i.e. the elements of the matrix I'.

The case P < <2P wifi now be considered. This is distinguished from the alternative by a check made after computing the compact SVD (step S2-4) in step S2-6.

In this case, a minor modification must be introduced to the reconstruction operation in equation (6) because the matrix F has only (12p)<p non-zero singular values and is (n-p)xp in size. Accordingly, must be extended with zeros before inserting it in equation (6), that is (in step S2-8): ( . t,U(2p_n)xp The geometrical explanation for this zero-padding is that the diagonal elements of are the principal cosines of the subspace Y with respect to the reference axes ( i If p«=n<2p, then 2p-n such principal cosines are equal to 1, hence (n-p)xp) 2p-n principal sines are equal to 0, which is the reason why the diagonal matrix containing the principal sines, ,, has to be extended with 2p -ii diagonal zeros.

Equations (5) to (7) can be conveniently used for conformance testing, to test if the compression method is implemented by a terminal. From the channel measurements, in the form of the H matrix, (or directly from the eigenvectors y1,...,y) and from the feedback information, it is straightforward to check if equation (7) is satisfied.

It should also be rioted that any conventional source coding technique can be used to further compress and quantise the matrix F, to generate the actual feedback bits: amongst the others, codebook-based vector/matrix quantisation or various forms of scalar quantisation of the matrix elements.

The above described compression scheme allows for reduction of the number of coefficients that are required to be encoded in order to feed back the spatial structure of a MIMO channel, thus reducing the uplink overhead required for feedback in single-cell or multi-cell DL (downlink) MIMO. This is achieved with little added complexity (one small SVD operation) at both ends of the communications channel.

While the above matrix operations are expressed in a certain manner, predicating right multiplication in specific steps, it will be understood that mathematically an alternative approach can be developed which involves left multiplication. No implicit limitation is to be placed on the scope of the invention by virtue of this feature of the description.

The reader will appreciate that the foregoing is intended to illustrate an example of the invention, and no limitation on the scope of protection is to be imputed therefrom. The scope of protection sought is set out in the attached claims, which are to be read in the light of the description (but not limited thereby) with reference to the accompanying drawings.

Claims (13)

1. CLAIMS: 1. A method of compressing data for transmission across a communications channel, the data representing channel conditions for the communications channel in the opposite direction, comprising arranging the data as a matrix comprising a number of orthonormal vectors derived from a channel matrix, determining a singular value decomposition of a subset of the orthonormal matrix to generate matrices respectively of left and right singular vectors, the number of vectors in the subset being equal to the order of the vectors, and right multiplying the remainder orthonormal vectors not included in the singular value decomposition by a matrix product of the matrix of right singular vectors and the matrix of left singular vectors to generate a matrix of compressed data.
2. 2. A method in accordance with claim 1 and comprising deriving said orthonormal vectors from a channel matrix representing the channel conditions of the communications channel in the opposite direction from that intended for transmission of said compressed data.
3. 3. A method of decompressing compressed data, said data being arranged as a square matrix, comprising determining a singular value decomposition of the square matrix, constructing a matrix Y' comprising: ,, v1(i_) yt UIEV1t where V1 is a right singular value matrix resultant from said singular value decomposition, U1 is a left singular value matrix resultant from said singular value decomposition, and is a diagonal matrix resultant from said singular value decomposition.
4. 4. A method for compressing the representation of a set of channel eigenvectors, in a channel state information (CSI) feedback scheme, comprising jointly encoding a set of p n-dimensional eigenvectors, with n p, by using n-p coefficients per vector.
5. 5. A method in accordance with claim 4 wherein encoding comprises arranging the set of eigenvectors in an n x p orthonormal matrix, taking a singular value decomposition (SVD) of the top p x p block of the matrix, and multiplying the bottom (n-p) x p matrix block by the SVD output to determine a channel representation.
6. 6. A method of decoding a channel representation comprising reversing a method in accordance with any one of claims 1, 2, 4 or 5 to reconstruct an orthonormal n x p matrix from a (n-p) x p block.
7. 7. A wireless communications apparatus operable to receive a signal from which it is able to derive channel state information, and operable to emit a signal to convey said derivation of channel state information, and comprising channel state information compression means operable to compress said derivation of channel state information in accordance with any one of claims 1, 2, 4 or 5.
8. 6. A wireless communications apparatus operable to receive a signal conveying a derivation of channel state information, operable to decompress said signal in accordance with claim 3 or claim 6.
9. 9. A wireless communications apparatus in accordance with claim 8 and operable to generate a precoding scheme on the basis of received channel state information.
10. 10. A computer program product comprising computer executable instructions operable to configure a general purpose computerised communications device to perform a method in accordance with any one of claims I to 6.
11. 11. A computer program product in accordance with claim 10 and comprising a computer readable storage medium.
12. 12. A computer program product in accordance with claim 10 and comprising a computer receivable signal.
13. 13. A wireless communications system comprising a wireless communications apparatus in accordance with claim 7 in wireless communication with a wireless communications apparatus in accordance with claim 8 or claim 9.