The present invention relates generally to wireless telecommunications, and in particular to a method of compressing signals for transmission on backhaul channels in wireless telecommunication networks.
The exponential growth in the demand of wireless data communications has put a tremendous pressure on the cellular network operators to improve the capacity of their communication networks. To improve the spectral efficiency of these networks, scarce radio resources have to be reused aggressively in neighboring cells. As a result, inter-cell interference has become a main source of signal disturbance, limiting not only the service quality of the cell-edge users, but also the overall system throughput.
Coordinated multi-point (CoMP) transmission or reception is one known means to effectively mitigate inter-cell interference.
To enable the CP 30 to effectively coordinate transmission and/or reception at multiple cells 14, timely signal information must be communicated between remote base station sites 16 and the CP 30. However, the amount of information needed to send to or receive from each base station 16 can be overwhelming, especially when multiple antennas are deployed at each site. In general, the CP 30 generates, and must transmit to remote base stations 16, the In-phase (I) and Quadrature (Q) components of complex-valued downlink signals to be transmitted by each antenna at each base station 16. Additionally, each base station 16 must transmit to the CP 30 the complex-valued IQ signal received at each antenna.
In the standard Common Public Radio Interface (CPRI), each real-valued sample of the IQ backhaul signal would simply be quantized independently by a fixed number of bits (e.g., 16). CPRI transmission does not exploit any structure of the underlying backhaul signal, and is an inefficient way of representing wireless communication signals. CPRI transmission of IQ wireless signals thus places a large burden on the capacity of backhaul links, which may limit the performance otherwise achievable by CoMP systems.
Other systems may additionally perform precoding at a processing unit of a network, for transmission by base stations. All such systems require voluminous data transfer between the network processing unit and base stations.
According to one or more embodiments described and claimed herein, a compression/decompression method reduces backhaul bandwith required for communicating a complex-valued radio signal between base stations and a processing unit, such as for example the Central Processor of a Coordinated MultiPoint (CoMP) system. The method exploits both the spatial and temporal correlation of the wireless IQ signal in order to remove redundancy and substantially reduce the capacity requirement of the backhaul links. Feature component signals of significance are extracted through linear transformation from the radio signal, and are individually quantized, possibly at different bit rates in accordance with their relative importance. The transformation can either be pre-determined or computed in real-time based on the spatial and temporal statistics of the radio signal. In the latter case, the transformation matrix or matrices are also sent over the backhaul in order to allow the radio signal to be reconstructed at the receiving end. Different methods of generating the transformation matrices are proposed.
One embodiment relates to a method of transmitting wireless communication signals across a backhaul channel between a base station and a network processing unit. A wireless communication signal is represented as a plurality of time-domain, complex-valued signal samples. At least one linear transformation of the signal samples is performed across one or more of time and antenna spatial domains to generate transformed signal coefficients. The transformed signal coefficients are quantized. The quantized transformed signal coefficients are transmitted across the backhaul. Quantization information is also transmitted across the backhaul.
Another embodiment relates to a method of receiving wireless communication signals across a backhaul channel between a base station and a network processing unit. Quantized transformed signal coefficients are received across the backhaul. Quantization information is received across the backhaul. Transformed signal coefficients are decoded from the quantized transformed signal coefficients based on the quantization information. At least one inverse linear transformation of the transformed signal coefficients is performed across one or more of time and antenna spatial domains to recover a plurality of time-domain, complex-valued signal samples corresponding to a wireless communication signal.
Yet another embodiment relates to a base station. The base station includes one or more antennas, and at least one front end circuit operative to front end process a wireless communication signal received from the antenna, and generate a plurality of time-domain, complex-valued signal samples. The base station further includes at least one transformation unit operative to perform a linear transformation of the signal samples across one or more of time and antenna spatial domains to generate transformed signal coefficients. The base station also includes a transmit bit allocation unit operative to determine a number of quantization bits for quantization of the transformed signal coefficients based on the variance of the coefficients. The base station further includes a quantizer operative to quantize the transformed signal coefficients using the number of bits determined by the bit allocation unit. The base station additionally includes a backhaul communication interface operative to transmit quantized transformed signal coefficients and quantization information across the backhaul to a network processing unit.
Still another embodiment relates to a network processing unit. The processing unit includes a backhaul communication interface operative to receive quantized transformed signal coefficients and quantization information from a plurality of base stations. The processing unit also includes a receive bit allocation unit operative to determine a number of quantization bits for decoding quantized transformed signal coefficients received from a base station, based on the received quantization information. The processing unit further includes a decoder operative to decode the received quantized transformed signal coefficients based on a number of bits determined by the bit allocation unit. The processing unit additionally includes at least one inverse transformation unit operative to perform an inverse linear transformation of the transformed signal coefficients across one or more of time and antenna spatial domains to generate a plurality of time-domain, complex-valued signal samples. The processing unit also includes a processing circuit operative to process the signal samples.
According to embodiments of the present invention, backhaul communications between a network processing unit 30 and base stations 16 in a wireless communication system 10 are significantly reduced. To provide context, and solely for the purpose of explanation herein, embodiments of the invention are described wherein the network processing unit 30 is a Central Processor (CP) 30 of a Coordinated MultiPoint (CoMP) system 10. However, the invention is not limited to CoMP systems 10, but may find applicability in any wireless communication system 10 in which a network processing unit 30 performs signal processing, such as precoding, on signals to be transmitted by base stations 16, and which must be communicated to the base stations 16 over a backhaul communication link. Accordingly, all references herein to a CP 30 or a CoMP system 10 are to be understood as a non-limiting example of a network processing unit 30 in a wireless communication network 10.
The backhaul communications are reduced by performing targeted compression on complex-valued IQ radio signals prior to backhaul transmission, and corresponding decompression on the signals following backhaul reception. The signal compression may be applied to both downlink signals to be transmitted by base station 16 antennas, and uplink signals received by base station 16 antennas. The compression methodology extracts feature component signals of significance through linear transformations in time and/or antenna spatial domains, and individually quantizes them. The transformation can either be pre-determined or computed in real-time based on the spatial and temporal statistics of the radio signal.
The base station 16, also depicted in
In a typical wireless communication system, signals are transmitted in a multitude of time slots. Let T denote the duration of one such time slot. Within each time slot, the statistical behavior of the received or transmit signal remains roughly stationary, which is especially true when the base stations 16 in the network are synchronized. Let Ts denote the sampling time period of the radio signal, and let N≡T/Ts be the number of time samples within a time slot. Let y[n,m]=[y1[n,m],y2[n,m], . . . ,yn
According to embodiments of the present invention, the backhaul signal is compressed by exploiting its spatial and temporal correlation through linear transformations. In one embodiment, two stages of linear transformations are applied to the sampled signal y[n,m]. The structure of the encoding unit 100 and decoding unit 200 are depicted in
The linear transformation Wt 110, which is applied to the time-domain sampled signals, is described first. Let Yi[m]≡[yi[l,m],yi[2,m],L,yi[N,m]]T denote a vector formed by stacking the time samples received from, or to be transmitted by, antenna i in time slot m. For example, the time-domain analysis and synthesis transformation, Wt 110 and Tt 210 (see
Let the eigen decomposition of RY be UtΛtUtH, where ΛY is a diagonal matrix with the eigenvalues {λt,j}Nj=1 of RY as its diagonal elements, and Ut is an orthonormal matrix with the eigenvectors of RY as its columns. In this case, the analysis and synthesis transformation matrices 110, 210 should be set as Wt=UtH and Tt=Ut, respectively.
The intermediate samples, {zi[n,m]}, generated after the time-domain transformation 110 are related to Yi[m] for each antenna i through
The intermediate samples can be compressed using a-priori information, such as nulls in the intermediate transform domain created by scheduling. In this case, the compression unit 100 can determine which transformed intermediate transform domain components should be quantized and which components should be excluded. A similar method can be applied to exploit time-domain scheduling information in, for example, a time-division multiplexed system, in which case, Wt 110 may simply be chosen as the N by N identity matrix.
After the time-domain transformation 110, the intermediate samples are grouped across the spatial dimension, i.e. z[n,m]=[z1[n,m],z2[n,m],L,zn
The transformations in spatial domain 120 (i.e., exploiting correlation across antennas), are now described. Let Wa 120 and Ta 220 denote an analysis transformation matrix 120 of dimension Ka×na and a synthesis transformation matrix 220 of dimension na×Ka respectively. Since we are interested in compressing the signal, the transformation matrices are designed such that Ka≦na. For each time sample n in time slot m, a feature vector x[n,m] of dimension Ka is computed using Wa 120 given by
x[n,m]≡[x1[n,m],x2[n,m],L,xK
and then quantize 130 each component feature xi[n,m] individually using a codebook Ci of size |Ci|=2b
such that Ta{circumflex over (x)}[n,m] closely approximates z[n,m]. For example, the compression unit 100 should compute
From the orthogonality principle, we have
Hence, for any given synthesis transform Ta 220, the best analysis transform 120 is given by its pseudo inverse
Wa=(TaHTa)−1TaH, (3)
which projects z[n,m] onto the column space of Ta. Substituting equation (3) into (2), we have
In general, the quantization of all feature components should be performed jointly over the product codebook
with the joint distortion metric
d(x,y)=(x−y)HTaHTa(x−y)
to achieve the best performance.
In the special case where the columns of Ta 220 are chosen to be orthogonal, we have
where wi denotes the squared norm of the i-th column of Ta 220. Hence, the compression unit 100 can quantize 130 each feature component xi[n,m] using Ci individually with the per-feature distortion metric
d(x, y)=|x−y|2
without affecting the overall performance. This presents significant complexity reduction since joint optimization across all feature dimensions simultaneously is not required, as in the original formulation in equation (1).
In one embodiment of the spatial transform, Ta 120 is a square orthonormal matrix with Ka=na. In this case, also Wa=TaH. Ta 120 can be chosen to be a standard predetermined orthonormal matrix, such as the FFT or the DCT matrices. Alternatively, an orthonormal matrix can be adaptively computed that maximizes the energy compaction in fewer feature components. When the signal samples z[n,m] are Gaussian distributed, the best transform is the Karhunen-Loeve Transform (KLT) U whose columns are the eigenvectors of the signal covariance matrix Rz≡E[z[n,m]z[n,m]H], which can be computed approximately by averaging over different time samples and time slots as
Moreover, the eigenvalues {λi} of Rz also provides the relative significance of the corresponding feature component {xi[n,m]}, which can be used to determine the number of bits {bi} 140 used to quantize {xi[n,m]} 130. Specifically, λi=σi2, where σi2 denotes the variance of the i-th transformed coefficient. There are a number of possible methods to determine {bi} 140 depending on the type of quantizer 130 used. For fixed-rate quantization, {b1} 140 can be computed using a high-resolution approximation as
where btotal denotes the total number of bits available to index all components {{circumflex over (x)}i[n,m]}. The approximation can be further simplified by allocating equal number of bits to the first k components, where σk2>β and k<na. In this case, β is the minimum energy that determines if a feature component should be neglected. Alternatively, one can select the Breiman, Friedman, Olshen, and Stone (BFOS) algorithm to optimally allocate the bits for a given set of component codebooks {Ci}, as described in the paper by E. A. Riskin, “Optimal bit allocation via the generalized BFOS algorithm,” published in the IEEE Trans, Info Thy., vol 37, pp. 400-402, March 1991, the disclosure of which is incorporated herein by reference in its entirety. The quantizer 130 for each coefficient can also be a variable-rate quantizer 130. In this case, it is preferred to use a quantizer 130 with a fixed step or cell size. The fidelity of the reproduced signal is determined by the choice of the step or cell size.
In one embodiment, the spatial transformation matrices 120, 220 described above are computed in real time based on the spatial and temporal statistics of the radio signal, which are also measured from the radio signal. The coefficients of the spatial transformation matrices 120, 220 are transmitted over the backhaul, to allow the radio signal to be reconstructed at the receiving end. The frequency at which such statistics and coefficients are computed depends on the stationarity of the underlying signal, which depends on the scheduling decisions and user activity. For example, in the LTE network, this period lasts at least 2 time slots. The representation of the spatial transform coefficients and the bit allocation information can be done using fixed-rate (for example, 8 bit per coefficient) quantization 130.
Embodiments of the present invention provide an effective method to compress the digitized representations of complex-valued radio signals either received from or to be transmitted by each antenna at a remote base station 16 in a CoMP cluster 10. The compression exploits spatial and temporal correlations in the radio signals, and extracts feature components from these signals. In some embodiments the method utilizes a-priori knowledge, such as known scheduling information, to neglect part of the signal component and reduce both computational and backhaul bandwidth burden.
Embodiments of the present invention can optimally compress the feature components of wireless communication signals according to known total backhaul bandwidth constraints. The bandwidth is then allocated to each feature component according to its significance. Moreover, in some embodiments, because feature components are sorted according to their significance, the backhaul network may adaptively reject lower impact feature components, to further preserve bandwidth in the backhaul links.
The present invention may, of course, be carried out in other ways than those specifically set forth herein without departing from essential characteristics of the invention. The present embodiments are to be considered in all respects as illustrative and not restrictive, and all changes coming within the meaning and equivalency range of the appended claims are intended to be embraced therein.
Number | Name | Date | Kind |
---|---|---|---|
4694482 | Reesor et al. | Sep 1987 | A |
8169978 | Sheng et al. | May 2012 | B2 |
20040186867 | Schenk | Sep 2004 | A1 |
20060117078 | Hou et al. | Jun 2006 | A1 |
20070153731 | Fine | Jul 2007 | A1 |
20080025416 | Jafarkhani et al. | Jan 2008 | A1 |
20080095262 | Ho et al. | Apr 2008 | A1 |
20080120095 | Oh et al. | May 2008 | A1 |
20090082064 | Li et al. | Mar 2009 | A1 |
20090185607 | Lee et al. | Jul 2009 | A1 |
20090245399 | Lee et al. | Oct 2009 | A1 |
20090297054 | Regunathan et al. | Dec 2009 | A1 |
20100062775 | Hwang | Mar 2010 | A1 |
20100103834 | Gorokhov et al. | Apr 2010 | A1 |
20100198990 | Shimada et al. | Aug 2010 | A1 |
20100208834 | van Zelst et al. | Aug 2010 | A1 |
20110122957 | Tamaki et al. | May 2011 | A1 |
20120057535 | Zhang et al. | Mar 2012 | A1 |
Number | Date | Country |
---|---|---|
WO 9854850 | Dec 1998 | WO |
Number | Date | Country | |
---|---|---|---|
20120243468 A1 | Sep 2012 | US |