Embodiments of the present invention relate to the field of wireless device calibration; more particularly, embodiments of the present invention relate to calibrating transceivers in a wireless device based on transmissions with other transceivers in the wireless device.
RF-impairment calibration has been thought to be the major challenge in enabling reciprocity-based MU-MIMO. One internal self-calibration method, i.e., the first calibration method which does not require over the air feedback and can enable reciprocity-based MU-MIMO based on moderate and large arrays has been demonstrated.
Conventional downlink MU-MIMO schemes have been at the forefront of investigations in the past decade. These schemes promise spectral efficiency increases by using multiple antennas at the base-station and serving multiple users simultaneously without the need for multiple antennas at the user terminals. This is achieved by using knowledge of the channel state information (CSI) between each user and the transmitting base-station. Having CSIT (CSI available at the transmitter) allows the transmitter to precode the user-terminal streams so that each user terminal sees only its own stream. Given a base station with M transmit antennas, K single-antenna user terminals can be served simultaneously, giving roughly a multiplexing gain equal to min(M,K) with respect to a system serving a single terminal.
For the transmitter to achieve this operation reliably, the transmitter needs to have sufficiently accurate CSIT, i.e., the transmitter needs to know the channels between itself and each of the users sufficiently accurately. The techniques used for acquiring CSIT fall into two categories. The first category employs M pilots (one per base-station transmit antenna) in the downlink to allow each user terminal to estimate the channel coefficients between the user-terminal's own antenna(s) and those of the base-station. This operation provides each CSI at each receiving user-terminal (CSIR) regarding the channel between each base-station transmit antenna and the user-terminal receive antennas. The CSIR, i.e., the CSI information available at each user-terminal, is then fed back to the transmitter by use of uplink transmissions to provide CSIT, i.e., CSI at the transmitting base-station. This category of CSIT acquisition schemes have two overheads: (i) a downlink pilot overhead, which scales linearly with M (then number of antenna elements at the transmitting base-station); (b) an uplink feedback overhead, responsible for making available to the base-station the channels between each user-terminal and each base-station antenna. In the case, each user terminal has a single antenna, and the uplink feedback is responsible for providing to the base-station the MK channel coefficients (complex-scalar numbers), one coefficient for each channel between each user terminal antenna and each base-station antenna. Although the uplink overhead could in principle be made to grow linearly with min(K,M), with the methods used in practice, this overhead grows as the product of M and K. The downlink overhead limits the size of the antenna array, M, that can be deployed. Similarly, the uplink overheads limit both M and K, as the overheads grow very fast with respect to increasing M and K.
The second category of CSIT acquisition techniques is referred to herein as reciprocity-based training schemes. They exploit a property of the physical wireless channel, known as channel reciprocity, to enable, under certain suitably chosen (M,K) pairs, very high-rate transmission with very efficient CSIT training. In particular, pilots are transmitted in the uplink by each user (K pilots are needed, but more could be used) and the corresponding pilot observations at the base-station are directly used at to form the precoder for downlink transmission. If the uplink training and the following downlink data transmission happen close enough in time and frequency (within the coherence time and the coherence bandwidth of the channel) then, the uplink training provides directly the required (downlink channel) CSI at the transmitter, since the uplink and the downlink channels at the same time and frequency are the same. In this category of techniques, the uplink overheads scale linearly with K, i.e., with the number of user terminals that will be served simultaneously. These schemes are also typically envisioned as relying on TDD (Time Division Duplex) in order to allow uplink training and downlink transmission within the coherence bandwidth of the user terminal channel with a single transceiver shared for uplink and downlink data transmission.
One attractive aspect of reciprocity-based training schemes is that one can keep on increasing the size of the transmit antenna array, M, making it “massive”, without incurring any increase in the training overheads. Although with M>K, increasing M does not increase the number of simultaneously multiplexed streams, K, (i.e., K streams are simultaneously transmitted, one to each user), increasing M induces significant “beamforming” gains on each stream (which translate to higher rate per stream), at no additional cost in training. Alternatively, increasing M allows reducing the transmit power required to yield a target rate to a user terminal, thereby allowing for greener transmission schemes.
The challenge with reciprocity based training schemes is that the “compound” uplink and downlink channels at the same time and frequency are not the same. Specifically, although the uplink and downlink physical channel components are the same, each compound channel between a “source node” (responsible for transmitting an information-bearing signal from the transmit antenna) and a destination node (attached to the receive antenna) includes additional impairments due to the transmitter (the circuitry, at the transmitter) and the receiver (the circuitry, at the transmitter). When the transmitter and receiver roles are interchanged, different impairments occur at each node, thereby rendering the two compound channels non-reciprocal.
However, these transmitter/receiver impairments vary or drift, very slowly with time (seconds to minutes to longer, depending on the quality of the circuitry). As a result, this opens the door for infrequent calibration, as a method to compensate for these transmitter/receiver impairments and bring reciprocity based MU-MIMO to fruition.
Consider the problem of enabling MU-MIMO transmission from an array of M transmit antennas to K single-antenna user terminals. The downlink (DL) channel between the i-th base-station transmitting antenna and the j-th user terminal is given by
{right arrow over (yji)}={right arrow over (rj)}{right arrow over (hji)}{right arrow over (ti)}{right arrow over (xi)}+{right arrow over (zji)}
where {right arrow over (xi)}, {right arrow over (hji)}, {right arrow over (yji)}, {right arrow over (zji)}, denote the transmitted signal from base-station antenna i, the DL channel between the two antennas, the observation and noise at the receiver of user terminal j, respectively. The scalar (complex) coefficient {right arrow over (rj)} contains the amplitude and phase shifts introduced by RF-to-baseband conversion hardware (gain control, filters, mixers, A/D, etc.) at the receiver of user terminal j. Similarly, the scalar (complex) coefficient {right arrow over (ti)} contains the amplitude and phase shifts introduced by the baseband-to-RF conversion hardware (amplifiers filters, mixers, A/D, etc.) at the transmitter generating the signal to be transmitted by base-station antenna i.
Similarly the uplink channel between the j-th user terminal and the i-th base-station antenna is given by
where
denote the transmitted signal from user terminal j, the uplink (UL) channel between the two antennas, the observation and noise at the receiver of base-station antenna i, respectively. The scalar (complex) coefficient contains the amplitude and phase shifts introduced by RF-to-baseband conversion hardware (gain control, filters, mixers, A/D, etc.) at the receiver of base-station antenna i. Similarly, the scalar (complex) coefficient
contains the amplitude and phase shifts introduced by the baseband-to-RF conversion hardware (amplifiers filters, mixers, A/D, etc.) at the transmitter generating the signal to be transmitted by user terminal j.
In the uplink, we have the following model:
where is the vector of dimension Kx 1 (i.e., K rows by l column) comprising the user symbols on subcarrier n at symbol time t,
is the M×K channel matrix that includes the constant carrier phase shifts and the frequency-dependent constant in time phase shifts due to the relative delays between the timing references of the different terminals,
and
are the received signal vector and noise at the user terminals,
In the downlink, we have the following model
{right arrow over (y)}={right arrow over (x)}{right arrow over (T)}{right arrow over (H)}{right arrow over (R)}+{right arrow over (z)}
where {right arrow over (x)} is the (row) vector of user symbols on subcarrier n at symbol time t, {right arrow over (H)} is the K×M channel matrix that includes the constant carrier phase shifts and the frequency-dependent constant in time phase shifts due to the relative delays between the timing references of the different terminals, {right arrow over (y)} and {right arrow over (z)} are the received signal (row) vector and noise at the user terminals, {right arrow over (R)}=diag({right arrow over (r1)}, {right arrow over (r2)}, . . . , {right arrow over (rK)}) and {right arrow over (T)}=diag({right arrow over (t1)}, {right arrow over (t2)}, . . . , {right arrow over (tM)})
The matrices ,
{right arrow over (R)} and {right arrow over (T)} are unknown locally constant diagonal matrices. For purposed herein, “locally constant” means that they might vary over very long time (certainly, much longer than the coherence time of the channel), mainly due to thermal drift effects, but they do not depend on any “fast effects” such as frequency offsets, and propagation time-varying fading, since these effects are all already taken care of by the timing and carrier phase synchronization, and included in the matrices {right arrow over (H)} and
. By reciprocity of the physical channel, the following holds
{right arrow over (H)}=
For simplicity, the thermal noise is neglected. In order to estimate the downlink channel matrix, the K user terminals send a block of K OFDM symbols, such that the uplink-training phase can be written as
where is a scaled unitary matrix. Hence, the base-station can obtain the channel matrix estimate
However, in order to perform downlink beamforming we need the downlink matrix {right arrow over (T)}{right arrow over (H)}{right arrow over (R)}. While reciprocity ensures that the physical channel component in the uplink estimated channel yields immediately the corresponding component in the downlink channel (it is assumed that uplink training and downlink data transmission occur in the same channel coherence time), the transmit and receive diagonal matrices need to be known for the downlink, while the product of those matrices for the uplink and the channel matrix {right arrow over (H)}= are known, which are generally arbitrarily related.
As a prelude to describing one relative calibration method, notice that the downlink channel matrix {right arrow over (T)}{right arrow over (H)}{right arrow over (R)} is not entirely needed to perform beamforming. In fact, only the column-space of this matrix is needed; that is, any matrix formed by {right arrow over (T)}{right arrow over (H)}A, where A is some arbitrary invertible constant diagonal matrix, is good enough for any kind of beamforming. For example, consider Zero Forced Beamforming (ZFBF). The ZFBF precoding matrix can be calculated as
W=Λ
1/2
[A
H
{right arrow over (H)}
H
{right arrow over (T)}
H
{right arrow over (T)}{right arrow over (H)}A]
−1
A
H
{right arrow over (H)}
H
{right arrow over (T)}
H
where Λ is a diagonal matrix that imposes on each row of the matrix W, the row normalization ∥wm∥2=1, for all m. Hence, the ZFBF precoded signal in the downlink will be
We notice that the resulting channel matrix is diagonal, provided that K≦M. It follows that the problem is how to estimate {right arrow over (T)}{right arrow over (H)} up to the left multiplication by some known matrix A, from the uplink training observation knowing that {right arrow over (H)}=
. Following the relative calibration procedure of Shepard et al. in “Argos: Practical Many-Antenna Base Stations” (Mobicom 2012, pp. 53-64; hereinafter “Argos”), the fact that the diagonal matrices
,
, {right arrow over (R)}, and {right arrow over (T)} are essentially constant in time for intervals much longer than the slot duration can be exploited (the calibration procedure may be repeated periodically, every some tens of seconds or even more, depending on the hardware stability, temperature changes, etc.).
The procedure, amounting to one calibration method, consists of the following steps:
i.e., it is a diagonal matrix containing the coefficients due to the other base-station antennas receive RF chains, the (M−1)×1 vector hS←1 denotes physical channel from reference base-station antenna 1 to the rest of the base-station antennas, and the (M−1)×1 vector represents thermal noise at the (M−1) non-transmitting base-station antennas.
where is the coefficient due to the calibration-reference antenna receive RF chain.
is an irrelevant constant term that depends only on the calibration-reference antenna up and down modulation chains.
At this point, the desired downlink channel matrix can be obtained from the calibration matrix and the uplink estimated channel matrix
simply by multiplication with the uplink estimated channels. In particular, we have,
where A=α1
The self-calibration process of this calibration method takes at least M OFDM symbols, one symbol for the pilot from reference antenna to all other base-station antennas, and M−1 OFDM symbols to send orthogonal training sequences from all the other base-station antennas to the calibration-reference antenna.
This calibration method has its limitations. First, note that the relative calibration of each base-station antenna with respect to the reference antenna is formed as the ratio of two observations, and, in particular, by dividing
The noise in the dividing term [yS→1]m-1, can cause a large estimation error in the calibration estimate. This effect was indeed noticed by the developers of this calibration method: “Another challenge we encountered while performing our indirect calibration approach is the significant amplitude variation for the channels between the reference antenna 1 and other antennas. This is due to the grid-like configuration of our antenna array where different pairs of antennas can have very different antenna spacings. According to our measurement, the SNR difference can be as high as 40 dB, leading to a dilemma for us to properly choose the transmission power for the reference signal.” Their solution was to carefully place the reference antenna with respect to the rest of the base-station antennas: “we isolate the reference antenna from the others, and place it in a position so that its horizontal distance to the other antennas are approximately identical. Such placement of the reference antenna does not affect the calibration performance due to our calibration procedure's isolation of the radio hardware channel from the physical channel.”
Such a need for careful placement of the reference antenna with respect to the rest of the transmitting antennas is a significant limiting factor in deployments relying on this calibration method, which significantly limits their efficacy in downlink MU-MIMO deployments from sets of non-collocated antennas.
In general, for noise robustness purposes, much larger blocks and maximal ratio combining of the received power can be used, such that we sent D pilot symbols from reference antenna 1 to the other base-station antennas, and M−1 orthogonal training sequences over (M−1)D symbols from the other base-station antennas to the reference one, achieving a factor D in signal to noise ratio for calibration, where D≧1 is some sufficiently large integer in order to improve performance. However, this does not eliminate the inherent limitations of the Argos calibration methods especially for scalable and distributive deployments.
A method and apparatus is disclosed herein for internal relative transceiver calibration. In one embodiment, the method comprises generating a plurality of processed observations corresponding to pilots transmitted by transceiver units in the entity and observations of the pilots by transceiver units in the entity, each processed observation being indicative of a combined response of a pilot transmitted by transmitter hardware of one transceiver unit at the entity and an observation of the pilot by receiver hardware of another transceiver unit at the entity; grouping the plurality of processed observations into one or more observation pairs, where each observation pair of the one or more observation pairs comprises: a first observation indicative of a combined response between transmitter hardware of a first transceiver unit and receiver hardware of a second transceiver unit in the entity, and a second observation indicative of a combined response between transmitter hardware of the second transceiver unit and receiver hardware of the first transceiver unit in the entity; and calculating, based on at least one of observation pairs, relative calibration values, where each relative calibration value is associated with a transceiver of each transceiver unit and is relative with respect to a transceiver of a reference unit at the entity.
The present invention will be understood more fully from the detailed description given below and from the accompanying drawings of various embodiments of the invention, which, however, should not be taken to limit the invention to the specific embodiments, but are for explanation and understanding only.
a)-(c) illustrate three examples of pairs of measurements that can be used for self-calibrating a circular array of antenna elements.
cl,
cl), associated with the two-way measurements shown in
Embodiments of the invention include methods and apparatuses that allow distributed readily scalable relative calibration methods. These relative calibration methods can be used for providing calibration that robustly enables wireless communication schemes, such as, for example, high-performance reciprocity-based downlink MU-MIMO schemes. In particular, the disclosed calibration methods enable MU-MIMO schemes that utilize small, large, or Massive MIMO arrays, with collocated or non-collocated antenna elements. Examples of the non-collocated case involve network MIMO in cellular, transmission from remote radio heads (RRH), but also more general MU-MIMO schemes, in which user terminals are simultaneously served by different (overlapping) sets of antennas in a field of antenna elements.
Embodiments of the invention included a combination of new reference-signaling methods for calibration and new techniques for performing calibration, enabling resource-efficient and reliable and robust calibration for network Massive MIMO, MU-MIMO based on remote radio heads, hierarchical calibration, as well as on demand, distributed calibration for reciprocity based MU-MIMO based on set of possibly overlapping arrays of non-collocated antenna elements.
Embodiments of the invention have the following advantages with respect to other approaches.
In the following description, numerous details are set forth to provide a more thorough explanation of the present invention. It will be apparent, however, to one skilled in the art, that the present invention may be practiced without these specific details. In other instances, well-known structures and devices are shown in block diagram form, rather than in detail, in order to avoid obscuring the present invention.
Some portions of the detailed descriptions which follow are presented in terms of algorithms and symbolic representations of operations on data bits within a computer memory. These algorithmic descriptions and representations are the means used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. An algorithm is here, and generally, conceived to be a self-consistent sequence of steps leading to a desired result. The steps are those requiring physical manipulations of physical quantities. Usually, though not necessarily, these quantities take the form of electrical or magnetic signals capable of being stored, transferred, combined, compared, and otherwise manipulated. It has proven convenient at times, principally for reasons of common usage, to refer to these signals as bits, values, elements, symbols, characters, terms, numbers, or the like.
It should be borne in mind, however, that all of these and similar terms are to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. Unless specifically stated otherwise as apparent from the following discussion, it is appreciated that throughout the description, discussions utilizing terms such as “processing” or “computing” or “calculating” or “determining” or “displaying” or the like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system memories or registers or other such information storage, transmission or display devices.
The present invention also relates to apparatus for performing the operations herein. This apparatus may be specially constructed for the required purposes, or it may comprise a general purpose computer selectively activated or reconfigured by a computer program stored in the computer. Such a computer program may be stored in a computer readable storage medium, such as, but is not limited to, any type of disk including floppy disks, optical disks, CD-ROMs, and magnetic-optical disks, read-only memories (ROMs), random access memories (RAMs), EPROMs, EEPROMs, magnetic or optical cards, or any type of media suitable for storing electronic instructions, and each coupled to a computer system bus.
The algorithms and displays presented herein are not inherently related to any particular computer or other apparatus. Various general purpose systems may be used with programs in accordance with the teachings herein, or it may prove convenient to construct more specialized apparatus to perform the required method steps. The required structure for a variety of these systems will appear from the description below. In addition, the present invention is not described with reference to any particular programming language. It will be appreciated that a variety of programming languages may be used to implement the teachings of the invention as described herein.
A machine-readable medium includes any mechanism for storing or transmitting information in a form readable by a machine (e.g., a computer). For example, a machine-readable medium includes read only memory (“ROM”); random access memory (“RAM”); magnetic disk storage media; optical storage media; flash memory devices; etc.
A method and apparatus for calibrating transceiver units in an entity are described. In one embodiment, each unit includes an antenna element for use in joint transmission from the entity to at least one wireless entity. In one embodiment, the method comprises transmitting at least one pilot from each transceiver unit in the entity using its antenna element; receiving one or more observations at each transceiver unit in the entity using its antenna, each of the one or more observations corresponding to receipt of the at least one pilot being transmitted from one or more other transceiver units in the entity; generate a plurality of processed observations, each processed observation being indicative of a combined response between transmitter hardware of one transceiver unit at the entity and receiver hardware of another transceiver unit at the entity; grouping the plurality of processed observations into one or more observation pairs, where each observation pair of the one or more observation pairs comprises a first observation indicative of a combined response between transmitter hardware of a first transceiver unit and receiver hardware of a second transceiver unit in the entity, and an second observation indicative of a combined response between transmitter hardware of the second transceiver unit and receiver hardware of the first transceiver unit in the entity; and calculating, based on at least one of observation pairs, relative calibration values, where each relative calibration value is associated with a transceiver of each transceiver unit and is relative with respect to a transceiver of a reference unit at the entity.
In one embodiment, the calibration process is performed by a base station, access point or other type of wireless apparatus. In one embodiment, the base station comprises: a plurality of transceivers, each transceiver of the plurality of transceivers comprising an antenna element; a calibration processor coupled to the plurality of transceivers. The calibration processor is operable to control the antenna element of each transceiver unit to transmit at least one pilot; to control the antenna elements of the plurality of transceiver units to receive one or more observations, each of the one or more observations received by one transceiver unit corresponding to receipt of the at least one pilot being transmitted from one or more other transceiver units in the plurality of transceivers; to generate a plurality of processed observations, each processed observation being indicative of a combined response between transmitter hardware of one transceiver unit in the plurality of transceivers and receiver hardware of another transceiver unit in the plurality of transceivers; to group the plurality of processed observations into one or more observation pairs, where each observation pair of the one or more observation pairs comprises a first observation indicative of a combined response between transmitter hardware of a first transceiver unit and receiver hardware of a second transceiver unit of the plurality of transceivers, and a second observation indicative of a combined response between transmitter hardware of the second transceiver unit and receiver hardware of the first transceiver unit; and to calculate, based on at least one of observation pairs, relative calibration values, where each relative calibration value is associated with a transceiver of each transceiver unit and is relative with respect to a transceiver of a reference unit at the base station.
In one embodiment, the relative calibration values comprise estimates of relative calibration coefficients between the transceiver of each of the plurality of transceiver units and the transceiver of the reference unit. In one embodiment, a relative calibration coefficient of one of the transceiver units comprises the ratio of a calibration coefficient of the one transceiver unit over the ratio of a calibration coefficient of the reference unit, and where the calibration coefficient of the one transceiver unit is defined as the ratio of transmitter gain of the one transceiver unit over receiver gain of the one transceiver unit. In one embodiment, the estimates of the relative calibration coefficients are based on the effect of their values on an error metric, wherein the error metric is a function of individual error quantities, each individual error quantity based on a function of the two observations in an individual observation pair. In one embodiment, the relative calibration coefficients are selected as values that minimize the error metric, wherein the error metric is a sum of the squares of the individual error quantities, and wherein each of the individual error quantities is based on an individual pair is a linear combination of the observations in the pair. In one embodiment, the relative calibration values are calculated based on prior calibration estimates.
In one embodiment, the base station determines a channel matrix for a transceiver of at least one of its transceiver units using the estimates of relative calibration coefficients and then determines a precoder based on the channel matrix. The precoder is subsequently used to transmit data.
Referring to
In one embodiment, processor 220 provides parallel output symbols streams to modulators, MODS (230a through 230t). Each modulator 230 further processes (e.g., convert to analog, amplify, filter, and upconvert) the output sample stream to obtain a downlink signal. The downlink signals from modulators 230a through 230t are transmitted via antennas 235a through 235t, respectively.
In one embodiment, at base station 200, the uplink signals from various UE's or by other antennas, collocated at the same base station 200 or located at different base-stations or other wireless entities are received by antennas 235a through 235t, demodulated by demodulators (DEMODs 230a-230t). In one embodiment, the demodulated signals are detected by MIMO detector 240 and further processed by a receive processor 245 to obtain decoded data and control information sent by UE's and other wireless entities. In one embodiment, receive processor 245 receives detected signals from MIMO detector and provides decoded data to a data sink 250 and control information to the controller/processor 270. In one embodiment, the demodulated signals output by DEMODs 230a through 230t are also provided to the channel processor 280 where uplink channel may be estimated and provided to the controller/processor 270.
Base station 200 also includes a calibration-processing unit/processor 290. This is responsible for estimating (and possibly compensating for) the impairments introduced by RF-to-baseband conversion hardware (gain control, filters, mixers, A/D, etc.) coupled with each antenna element 235a through 235t when base station 200 processes uplink received signals from these elements, as well as the impairments introduced by the baseband-to-RF conversion hardware (amplifiers filters, mixers, A/D, etc.) coupled with each antenna element 235a through 235t when base station 200 generates the signals that are to be transmitted by base station antenna elements 235a through 235t. Viewing the combination of element 230a with element 235a as a single (non-calibrated) transceiver unit, and viewing all such combinations of elements 230a through 230t with their respective elements 235a through 235t as individual transceiver units, in one embodiment, calibration processor 290 performs processes for relative calibration of a subset of these transceiver units that are used to enable reciprocity based MU-MIMO from a subset of these transceiver units. In one embodiment, processor 290 exchanges control information with the controller/processor unit 270. The calibration processor 290 calculates calibration values, which are used at controller/processor 270 together with UL channel estimation to construct one or more precoding vectors for one or more UEs and provide them to TX MIMO processor 220 for precoding. In some embodiments, processor 290 is provided additional information from other base stations, indicative of signals received and/or transmitted by other base stations, to assist in relative calibration of transceiver units connected to separate base stations. The embodiments of this invention are enabled by processing unit 290, and involve both the signaling and data collection aspects of calibration as well as the relative calibration methods set forth herein, which are based on the collected data, and, possibly additional parameters, including past relative calibration values for arbitrary subsets of the transmit antenna nodes at this and possibly other base stations.
Controller/processor 270 directs the operation at the base station. Processor 270 and/or other processors and modules at the base station perform or direct operations and/or other processes for the techniques described herein. Memory 260 stores data and program code for the base station.
Consider an extension of the Argos approach involving the same topology, and the same number of calibration training slots, i.e. D slots per base station antenna (with D≧1). The extension is as follows: each antenna, including the calibration antenna 1, first broadcasts independently its training symbols. In one embodiment, this is performed using the same signaling dimensions as Argos, but the matrix {right arrow over (X)}calib is diagonal, i.e., when each of the antennas in the set S transmits the remaining set of antennas in S are not transmitting and thereby they can receive. After each antenna has broadcasted its training symbol(s), all the measurements are collected where the following equation
corresponds to the training symbol from antenna j to antenna i, for each i≠j, 0≦i, j≦M. This is in contrast to Argos which relies only on the set of observations yi1 and y1i, for all i. In the preceding equation, wij is an i.i.d. complex Gaussian noise sample, with appropriate variance (including the effect of the training length D, which may be a design parameter to trade-off efficiency for noise margin, as explained before). Since perfect physical channel reciprocity is assumed, the wireless channel in both directions is considered to be the same (i.e., hij=hji)
Grouping the above measurements in pairs, the following can be written:
where βij={right arrow over (t)}i{right arrow over (t)}jhij complex coefficients associated to the unordered pair of antennas i,j.
One goal is to estimate the relative calibration coefficients ci for i=1, 2, . . . , M, up to a common multiplicative non-zero constant. In the absence of noise, the following equality exists:
y
ij
c
i
=y
ji
c
i
=c
i
c
jβij.
In one embodiment, a natural cost function to be minimized is used in order to find the calibration coefficients is the sum of squared errors
Note that this cost function is an example of a (non-negative) error metric. This error metric is comprises the sum of individual error quantities, whereby each error quantity only depends on a single pair of antenna elements, i and j, and in particular, on pairs of observations collected by base station antenna i(j) during the transmission of a pilot from base station antenna element j(i). In the absence of noise, the error metric takes its minimum value of zero when the correct calibration coefficients are used in the calculation of the error metric. In one embodiment, given that these calibration coefficients are unknown, given a set of pairs of (noisy) calibration-pilot observations, calibration coefficient estimates are calculated by determining the values of the ci's that minimize the error metric.
Note that the above metric is very attractive because it allows for an efficient computation of the optimal calibration values. However, other such error metrics can be used, wherein the error metric is a function of individual error quantities, and where each individual error quantity is a function of observations between element i and j and the unknown coefficients ci and cj. In another embodiment, instead of a summation, a weighted summation of individual error quantities can be used. Alternatively, the summation is replaced with a “max” operator. In yet another alternative embodiment, logarithmic functions are used in front of individual error quantities in the summation. Another example of an error matrix corresponds to
and where Δ denotes some predetermined RSS threshold. This metric allows, in a very simple and systematic manner, the elimination of measurement pairs which are not received at sufficiently strong levels.
The set F defines the set of (i,j) pairs of ordered measurements (yij, yij) used for determining the calibration coefficients. In one embodiment, inorder to avoid the trivial all-zero solution, without loss of generality, the following is imposed: |c1|=1. Then, the calibration coefficients are found as the solution of the optimization problem:
Argos, corresponds to the case of the star topology, i.e., when we use only the measurements yi1 and y1i, for all i=2, 3, . . . , M, that is, F={(1,2), (1,3), (1,M)}. In this case, the objective function is given by
The minimum for fixed c1 can be obtained by differentiating with respect to ci* and treating ci, ci* as if they were independent variables in a manner well-known in the art (see Hjorungnes et al., “Complex-valued matrix differentiation: Techniques and key results”, IEEE Trans. on Signal Processing, 2007, Vol. 55, No. 6, pp. 2740-2746), and then setting all derivatives equal to zero. The following set of equations is obtained:
If interested in minimizing J, this solution can be replaced back into J and then solved for c1 in the form c1=exp(jθ1). This amounts to a line search over the phase interval [0, 2π), which can be easily handled numerically. However, in one embodiment, only the coefficients ci up to a common multiplicative constant are a concern, and thus this last step is not needed, and c1 can be set equal to 1 arbitrarily, but without loss of generality for the relative calibration purpose. Not surprisingly, as a special case, the same estimator advocated in Argos and reviewed in the previous section can be obtained.
Given a set of ordered pairs, F, the solution to the minimization problem, i.e., determine the calibration coefficients c1, c2, c3, . . . , cM, up to a multiplicative constant, can be readily obtained. First, given a set F, the set of unordered pairs can be determined as
F
o={all pairs(i,j); such that(i,j) is in F or (j,i) is in F}.
For example, in the case of the Argos star topology,
F
o={(1,2),(1,3), . . . , (1,M),(2,1),(3,1), . . . , (M,1)}
For convenience, the (F-dependent) set of preprocessed measurements is defined as follows:
The minimum for fixed c1 can be obtained by differentiating with respect to ci* and treating ci, ci* as if they were independent variables [4], and then setting all derivatives equal to zero, i.e.,
By expanding the partial derivative in the i-th equation above, the following equation is obtained:
where c and ai are both M-dimensional column vectors, where the j-th of vector c is given by cj, and where the j-th entry of ai is given by
As a result, the set of equations for i=1, 2, . . . , M can be described as in matrix notation as A c=0, where
Above, b1 is a column vector of dimension M. Its k-th entry contains the negative of the first entry of the vector ak. Similarly, the matrix Ãt has M rows and M−1 columns. Its element in the k-th row and m-th column is given by the (m+1)-th element of vector ak. Letting {tilde over (c)}1 denote the vector of all calibration coefficients in c without c1, i.e.,
{tilde over (c)}
1
=[c
2
c
3
. . . c
M]
The least-squares calibration solution {tilde over (c)}1 is given by the solution to the set of equations Ã1{tilde over (c)}1=b1c1, which is given by
Note that other variations of estimates based on the matrix A are possible. In one embodiment, the set of coefficients are chosen that minimize J(c1, c2, . . . , cM) subject to the average of the powers of the ci's equaling some predetermined value. In this case, the chosen vector of coefficients corresponds to the eigenvector of the matrix A associated with the smallest-magnitude eigenvalue of A (suitably scaled so as to have its average power equal to the predetermined value).
The choice of the set of pairs of measurements, F, used in the calibration problem, i.e., in choosing the set of measurement pairs that are non-zero, can greatly affect performance. Consider the example in
Referring to
a) shows the set F used by the Argos calibration method. Due to their proximity, calibration of reference-element 1 with respect to reference-element 2 can be expected to be fairly accurate. However, calibration of the reference-element with respect to elements far from it, e.g., element 7 can be expected to be much less reliable. Such problems were indeed noticed in Argos, and required careful placement of the calibration reference antenna with respect to all other antennas. Based on this, the Argos self-calibration technique would have poor performance and would not be sufficient for calibrating arrays of non-collocated antenna elements for Massive MIMO deployments.
b) shows calibration with based on the measurement pair set F={(1,2), (2,3), . . . , (5,6), (7,8), (8,9), . . . , (11,12), (1,12)}. Although the same number of measurements is used in both
Including also the pair (6,7) in F, i.e., including also the pair depicted by the dashed double arrow in
c) shows another example in which, with exactly the same wireless signaling overhead for calibration, about twice as many measurement pairs are used (in particular, for each antenna, pairs with the closest 4 nodes are used), yielding further performance improvements.
Referring to
Next, in one embodiment, processing logic selects a subset F of these pair-wise measurements for calibration (processing block 340). In one embodiment, the set F used for calibration includes a chosen subset of pair-wise measurements, out of all the possible (i,j) antenna-element pairs. In one embodiment, the set F only includes measurements exceeding a predetermined threshold in received power level. In one embodiment, a pair is included in F, if the minimum of the received power levels in the received pair of measurements exceeds a predetermined level. In one embodiment, other criteria are used to include or exclude a pair from the set F including limiting the total number of pairs of observations used, while guaranteeing that all the terms can be estimated, as e.g., illustrated in the examples shown in
Returning to
At processing block 350, processing logic computes relative-calibration values based measurements of the chosen subset of pairs in the chosen subset to calculate relative calibration values with respect to a reference antenna. In one embodiment, prior calibration estimates can be used as well.
Based on a given set of estimates, −ĉk of the relative calibration parameters, the MU-MIMO training and signaling operation may be performed. First, observations are collected based on uplink pilots. These observations are then used at a central server to obtain an MMSE estimate of the Hup, namely, Ĥup. Assuming the set of {ĉk}k=2N
If ĉi is replaced with ci/c1, and Ĥup with Hup, the matrix Halt takes the desired form with D diagonal. Consequently, given Halt, and any given precoder function V=V(Halt), such as e.g., ZFBF in V(Halt)=Λ1/2[HaltHHalt]−1HaltH, the effective downlink channel is given by Y=uVTB{tilde over (R)}+Z, with effective NU×NU channel matrix given by Φ=VTB{tilde over (R)}. Then the instantaneous rate of the quantity log2(1+SINRi) can be used as the performance metric for the ith user, with SINRi computed in the usual manner.
In one embodiment, calibration occurs in stages. In particular, consider the example in
Although the total set of Mm+Mn=30 array elements can be calibrated together with a single calibration LS step, in one embodiment, the calibration is performed in stages. In one embodiment, first intra-cluster calibration, i.e., calibrate each cluster separately and independently is performed, and then inter-cluster calibration is performed. Effectively, for intra-cluster calibration, only measurement pairs within a cluster are used, and due to the relative proximity of nodes within a cluster as opposed to those in different clusters, such intra-cluster pairs of measurements may be expected to be more reliable (higher SNR) for calibration. As a result, it is assumed that self-calibration in cluster m has yielded coefficients ĉi,m, satisfying ci,m≈αmĉi,m, for some unknown complex-valued scaling constant αm, and self-calibration in cluster n has yielded coefficients ĉj,n satisfying cj,n≈αnĉj,n, for some unknown complex-valued scaling constant αn. Thereafter, inter-cluster calibration effectively amounts to finding the set of cluster calibration coefficients for ci,m and cj,n up to a common scaling constant. This is equivalent to finding αm and αn, up to a common scaling constant, or equivalently αm/αn, or equivalently estimating the complex constant β in the relationship αm=βαn. This can be done via LS calibration based on any non-zero subset of inter-cluster measurement pairs, i.e., bidirectional measurements for some ((i,n), (j,n)) pairs of antennas for some (i, j), and replacing ci,m and cj,n, with the expressions αmĉi,m and αnĉj,n, in the pair of equations describing the measurement pair. Equivalently, letting y(i,m)(j,n) and y(i,n)(i,m) denote the measurements at nodes (i,m) and (j,n) respectively, when calibration pilots are sent by nodes (j,n) and (i,m) respectively, in one embodiment, the following optimization problem is used:
Letting ŷ(i,m)(j,n)=y(i,m)(j,n)ĉ(j,n), and ŷ(j,n)(i,m)=y(j,n)(i,m)ĉ(i,m), the problem can be recast as follows
This problem has exactly the same structure as the original intra-cluster LS problems. In particular, subject to a given set of inter-cluster measurement pairs defined by F′, the least-square calibration problem can be solved using the same steps as the ones used to solve the original least squares problem. In one embodiment, inter-cluster calibration is performed over more than two clusters by solving for a set of reference inter-cluster calibration coefficients in a similar manner.
In general, hierarchical calibration can be used to jointly calibrate groups of transceiver units, where the number of groups jointly calibrated is larger than 2. One embodiment involving such use of hierarchical calibration, involves limiting the complexity to calibrating groups with no more than “X” groups of units at time. With X=9, for instance, in one embodiment transceiver units would first be calibrated in groups of size, e.g., 9, (i.e., a need to limit the size of the matrix A based on which the calibration algorithm is performed), allowing each group of 9 transceiver units to be independently calibrated relative to their own reference unit. Hierarchical calibration can then be readily used to jointly calibrate much larger sets, without violating this “size” constraint. In one embodiment, first close-by sets of antennas of size 9 are calibrated based on the base-line algorithm (intra-cluster calibration). Then, (inter-cluster) calibration of 9 such closely located (already calibrated 9-element) sets can be performed, yielding calibrated groups of 92 elements. Then inter-cluster calibration of 9 such closely located (already calibrated 92-element) sets can be performed, yielding calibrated groups of 93 elements), etc. The complexity of each one of these set-of-9 calibration operations is fixed; it is the same as the calibration complexity of the basic relative calibration of a group of 9 un-calibrated antenna elements.
More specifically, for convenience cluster nodes within each cluster are re-indexed, and the m-th node in cluster i is denoted by (i,m). Let ci,m=Ri,m/Ti,m denote the unknown parameter of interest. It is assumed that for each i, sufficiently accurate intra-cluster calibration has been performed using an algorithm so that each node has been calibrated relative to a reference node in cluster ci. In particular, given that the algorithm applied to cluster i has returned intra-cluster calibration estimates {ĉi,m}m, we have
c
i,m
≈ĉ
i,m
c
i. (16)
for some unknown parameter c1. Also let Y(i,m)→(j,n) denote the observation at node (j, n) based on a pilot transmitted by node (i, m), i.e., an observation of the form (11), with i and j replaced by (i, m) and (j, n), respectively.
Consider clusters as nodes on a graph. Assume a connected cluster-network graph (cl,
cl), pilot bursts are transmitted and received by BSs across clusters over a connected spanning subgraph (
cl,
cl), including all the clusters and a subset of links (
cl⊂
). Also let
cl denote the set of undirected edges corresponding to
. A pair (i, j)∈
, if there is at least one pair of observations {Y(i,m)→(j,n), Y(j,m)→(i,n)} of the form (11) that are to be used for calibration, and where the pair is due to a pair of calibration pilots transmitted by BSs (i, m) and (j, n) on distinct OFDM symbols but within the coherence-time of the channel. Let
ij denote the set of all (m, n) index pairs for which such bi-directional pairs of observations are available between APs (i, m) and (j, n), in clusters i and j respectively. Thus, (i, j)Σ
if and only if the set
ij is non-empty.
A visual interpretation of the hierarchical calibration problem is shown in
cl,
cl), associated with the two-way measurements shown in
The solution can be readily derived by following the same steps. Letting
{tilde over (Y)}
(i,m)→(j,n)
=ĉ
(i,m)
Y
(i,m)→(j,n)
and using (20) we can re-express Jh, as a function of the ci's as follows
Letting {tilde over (c)}=(c2, . . . , cNcl|, the vector {tilde over (c)} that minimizes the above formula as a function of c1 is given by (16), and where A is the NC×NC matrix with element in its i-th row and j-th column given by
Notice that for some i≠j, the coefficient Ai,j may be zero, if ij is empty, i.e., if (i,j)∉
.
Referring to
After pairing measurements, processing logic selects a subset of these pairs of observations for inter-cluster calibration (processing block 540). In one embodiment, all the pairs are used for inter-cluster calibration. In another embodiment, only the pairs with sufficiently large RSS (see paragraph 49 above) in the pair are chosen in the set F.
After selection, processing logic uses the selected pairs of observations for computing the inter-cluster calibration adjustment coefficients (processing block 550). In one embodiment, the inter-cluster calibration comprises setting αn=1 (thus, not making any adjustments to the calibration coefficients of cluster n), computing a value for αm via the least-squares method outlined in the preceding paragraphs, and adjusting the calibration coefficients for each of the antennas in cluster m using the computed value of αm (i.e., setting the new calibration value for a given antenna in cluster m as its previous value times the inter-cluster adjustment αm).
In one embodiment, inter-cluster calibration is performed on demand. In one embodiment, intra-cluster calibration has already been performed at a previous time instance (e.g., it has already been used to serve users via MU-MIMO transmissions separately emanating from each of the cluster of arrays). As a UE (e.g., a mobile) passes through an area where it is advantageous to be served by a joint transmission through multiple sites, inter-cluster calibration is activated to calibrate the multiple antenna clusters. In one embodiment, first calibration pilots are transmitted by one (or more) of the antennas of one or more clusters. In response, each of the other clusters transmits back calibration pilots from a one or more of their antennas. In one embodiment, these antennas are chosen based on the received signal strength at these antennas from one or more first calibration pilots emanating from different clusters. In one embodiment, several layers of hierarchical calibration are applied, yielding successively larger clusters of relatively calibrated antenna arrays.
The above calibration techniques can be tailored towards calibrating cellular architectures for reciprocity-based MIMO transmission.
Consider signaling and calibration for a cellular network of the form shown in
In one embodiment, depicted in
Referring to
Next, collected observations from adjacent base stations are passed to the corresponding controller for each adjacent base station pair (processing blocks 640Ak, Bk, Ck, and processing blocks 650Ak, Bk, Ck) to be used in inter-calibration. Then controllers perform inter-calibration procedure as described in
In one embodiment, first intra-cell calibration is performed, as the “intra-cell calibration measurements” are higher-SNR measurements. In one embodiment, hierarchical calibration is performed once intra-cell calibration is completed as described above with respect to hierarchical calibration. In one embodiment, only subsets of inter-cell measurements of sufficiently high received-signal strength are used in the inter-cell calibration step. These embodiments use at least RM dimensions to calibrate the whole network, with R at least as large as 3, but possibly larger R.
In the case that the array size becomes “Massive”, i.e., when M becomes large, the cost of RM dimensions for calibration becomes significant, and other more efficient methods that only require order-M dimensions for calibration. In one embodiment, reuse-1 pilot schemes are first used to accomplish intra-cell calibration. This is possible due to the proximity of antennas on the same base station with each other relative to the antennas at other base-stations.
Referring to
In the resource element over which a pilot is transmitted from a single antenna element at BS Ak (during this stage of reuse-1 pilot signaling), the remaining antenna elements on base station Ak pick up observations that can be used for intra-cell calibration with respect to the transmitting antenna element of base station Ak (since the transmission of this antenna element is received at much higher signal levels than interfering pilots simultaneously transmitted from other base stations). The same is true when each of the other antennas transmits from base station Ak. Although reuse 1 accomplishes intra-cell calibration, additional pilots are then needed for inter-cell calibration. For example, considering the controller of the {Ak, Bk} cluster, additional calibration pilots are required to estimate the single inter-cell calibration coefficient needed to calibrate the (cluster of) antennas at base station Ak with the (cluster of) antennas at base station Bk, assuming the intra-cell calibration coefficients are available to the controller (and are accurate) from the preceding intra-cell calibration cycle (which was based on reuse-1 calibration signaling).
As inter-cell calibration can be achieved by estimating only a single inter-cell calibration coefficient in each 2-BS cluster, it can be accomplished by use of a few pilot transmissions from each BS at sufficiently high reuse factors. Given D transmissions per BS with reuse R′, the total number of dimensions needed is M+DR′. The flow diagram of
In one embodiment, the intra-calibration signaling (depicted by processing blocks 730Ak, 730Bk, and 730Ck) occurs after the inter-calibration signaling. In one embodiment, intra-calibration signaling process occurs less (or more) frequently than inter-calibration signaling. In an alternative embodiment, the intra-calibration signaling in processing block 730Bk, uses different resources than in processing blocks 730Ak, and 730Ck. In one embodiment, the resources used for intra-calibration signaling at processing block 730Bk, are used at neighboring base-stations Ak and Ck for data transmission, with possibly different transmission powers than the one used for the intra-calibration pilots at base station Bk.
One set of embodiments involves the use of sequential signaling for hierarchical calibration. The following example illustrates the use of the techniques set forth herein.
Consider a scenario involving inter-cluster calibration of clusters m and n as shown in
Node “A” from cluster m transmits a pilot first, as shown in
In one embodiment, illustrated in
In one embodiment, illustrated in
In one embodiment, the controller terminates the signaling after the completion of the operation in
In one embodiment, tFinal(1)<tInitial(k) for k>1, that is, intra-cluster reference signaling in completed prior to inter-cluster reference signaling. In one embodiment, tFinal(k)<tInitial(1) for k>1, that is, inter-cluster reference signaling in completed prior to intra-cluster reference signaling. In one embodiment, intra- and inter-cluster reference signaling overlap in time.
In one embodiment, subsequent to transmission of the node “A” that is shown in
In one embodiment, the pilot is sent on a beam formed by using as beam weight at each of the two nodes the conjugate of the received signal, adjusted by the relative pre-calibration coefficient between elements “a” and “c” so as to induce beamforming at node A. Consider a scenario, such as e.g., in
Using this, the observation at node k, in response to the pilot from node 0, is given by
where hk0 denotes the physical channel between antenna element 0 and antenna element k. Then nodes 1 through K send pilots simultaneously and their superposition is received at node 0:
where w0 represents noise, and pk is the pilot transmitted by node k. Now, to find a0 where c0=a0c1, the following may be used:
This is because, ignoring noise, we have
In practice, to combat noise, the pilot weights can be chosen so as to maximize the receive SNR in the second channel (the simultaneous transmission of the K pilots). Ideally, pk would be picked to be proportional to the conjugate of hk0tk. Unfortunately that requires knowledge of the tk's. In practice, one can get close to this pilots choice, by picking pk proportional to the conjugate of “yk0 divided by ak” (i.e., we first pre-calibrate by dividing with ak, and then conjugate, and use the resulting coefficient as the pilot), since this quantity equals, in the absence of noise, hk0tk.
The two sets of transmissions in
In one embodiment, intra-calibration, [tInitial(1), tFinal(1)], can be done after inter calibration signaling i.e. interval [tInitial(2), tFinal(2)] followed by interval [tInitial(3), tFinal(3)].
Referring to
Upon receiving pilots, the second cluster chooses one or many antennas to transmit inter-calibration pilot signals (processing block 410n) and transmits the inter-calibration pilot signals. In one embodiment, this transmission occurs fast enough with respect to the initial pilot transmission, so that they are both within the coherence time of the channel between the two antennas. In an alternative embodiment, this transmission occurs and also a replica transmission of the initial pilot transmission occurs, with such timing that they are both within the coherence time of the channel between the two antennas. Besides collecting observations on the transmitted pilot signals over the air (processing block 430n), the second cluster also provides its own pilot observations to the first cluster as seen in processing block 420n. Although this step is in principle performed via a wired connection, it can also be provided via a wireless data transmission.
The first cluster, m, receives the pilot observations provided by the second cluster, n, (processing block 425m). Upon receiving pilot observations, the first cluster performs inter-cluster calibration and adjusts its calibration coefficients (processing block 500m). The first cluster also provides relative calibration adjustment to the second cluster (processing block 440m). In response to the cluster n inter-cluster relative calibration adjustment, processing logic of cluster m updates cluster n relative calibration coefficients (processing block 445n). In the embodiment shown in the figure, cluster n performs relative calibration of its RF chains with respect to cluster m. The calibration adjustment corresponds to computing a quantity αn in the manner described in [0082]-[0084]. The corresponding quantity αm in cluster m, is set to 1, since cluster m is treated as the reference cluster.
In one embodiment, processing blocks 410m, 430n, 425m, 420n, 430m, 410n can be repeated many times.
The following embodiments enable calibration of a large network of arrays by means of layered clustering and calibrations. In one embodiment, calibration involves a sequence of pilot transmission and calibration operations, resulting in generating sequentially larger clusters of calibrated arrays. In one example, first pilots are transmitted with reuse density d1, and pilot power P1, to train small isolated clusters of antenna; then a second set of pilots are transmitted with pilot power P2 (e.g., P2>P1) and density d2 (e.g., d2<d1). Measurements from the second set of pilots allow increasing the size of calibrated clusters. In one embodiment of the second reference-signaling cycle for calibration, at least one of the antennas transmitting the pilots is determined based on calibration outcomes from round 1. In one embodiment, Pk may be varied across the field. In one embodiment, pilot beams may be transmitted simultaneously from multiple antennas.
The following embodiments consider calibration “on demand.” In one embodiment, an additional step of calibration is triggered by a user terminal's uplink pilots, signifying that clusters, each cluster having available intra-cluster calibration coefficients (but with no inter-cluster calibration available), should be used for joint signaling to the user terminal. In one embodiment, this is determined by means of the RSS in the uplink at the receiving antenna elements at the time of the UL pilots. In one embodiment, user-terminal uplink pilots can be also used to trigger a refreshment procedure on intra or inter calibrations. In one embodiment, additional (helper) antenna nodes can be used to assist in the on-demand calibration procedure. These nodes maybe outside the serving cluster, and may have a very low RSS to the user terminal in question.
In embodiments described herein, the task of signaling and calibration of one or more clusters of antennas was presented as though it is controlled by a single processor entity. However, as the “helper” node suggests, helper nodes can enable other forms of distributed calibration. In one embodiment, different controllers engaged in calibrating their own clusters of possibly (but not necessarily) overlapping sets of antenna elements, exchange information providing “helper node” information with each other to assist in each other's calibration efforts. In one embodiment, this is cluster-driven, on demand, and is only be used in one direction.
Whereas many alterations and modifications of the present invention will no doubt become apparent to a person of ordinary skill in the art after having read the foregoing description, it is to be understood that any particular embodiment shown and described by way of illustration is in no way intended to be considered limiting. Therefore, references to details of various embodiments are not intended to limit the scope of the claims which in themselves recite only those features regarded as essential to the invention.
The present patent application claims priority to and incorporates by reference the corresponding provisional patent application Ser. No. 61/696,648, titled, “Method and Apparatus for Internal Relative Transceiver Calibration for Reciprocity-based MU-MIMO Deployments,” filed on Sep. 4, 2012.
Filing Document | Filing Date | Country | Kind |
---|---|---|---|
PCT/US2013/032299 | 3/15/2013 | WO | 00 |
Number | Date | Country | |
---|---|---|---|
61696648 | Sep 2012 | US |