Documente Academic
Documente Profesional
Documente Cultură
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/262677756
CITATIONS
READS
47
98
5 authors, including:
Qi Zhang
Kai-Kit Wong
4 PUBLICATIONS 52 CITATIONS
SEE PROFILE
SEE PROFILE
Michail Matthaiou
Queen's University Belfast
125 PUBLICATIONS 1,087 CITATIONS
SEE PROFILE
966
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 8, NO. 5, OCTOBER 2014
I. INTRODUCTION
1932-4553 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
ZHANG et al.: POWER SCALING OF UPLINK MASSIVE MIMO SYSTEMS WITH ARBITRARY-RANK CHANNEL MEANS
967
2Since
the physical size of the antenna array depends on the operating freat high frequencies (e.g., 60
quency, it can be very small even with large
GHz communications [27]).
968
The th element of
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 8, NO. 5, OCTOBER 2014
can be written as
(8)
where
is the th column of
plication, we further get
(9)
which can be further written as
where
denotes the th element of and
is the th column
of . Assuming an ergodic channel, the achievable uplink rate
of the th user is given by [12]3
(15)
The minimum mean-square-error (MMSE) estimate of
from
is [28]
(16)
(10)
where
get
(17)
where
. Noting that
, the entries of
are i.i.d. Gaussian random variables with zero-mean and unitdenote the channel estimation error
variance. Let
and be the linear receiver matrix which depends on . Then,
after linear reception, we have
(18)
As such, the estimated signal for the th user is given by
(12)
where
(19)
where
is the
rewritten as
, where
which satisfies
and
is the transmit pilot power. As a result, the BS
noisy pilot matrix as
receives the
where
and are the th columns of and , respectively.
Note that the last three terms in (20) correspond to intracell interference, channel estimation error and noise, respectively. According to the classical assumption of worst-case uncorrelated
Gaussian noise [12], along with the fact that the variance of elements of estimation error vector is
(13)
3Hereafter, the subscript will correspond to the case of perfect CSI, while
the subscript to the imperfect CSI case.
4Since the LOS channels are hardly changing, the BS may estimate the LOS
components during the previous transmission from the users to the BS. While
factor in massive MIMO systems is an interthe estimation of the Ricean
esting topic for additional research, we do not pursue it herein due to space
constraints.
th column of
(20)
(21)
the achievable uplink rate of the th user can be given by (22)
(at the bottom of the page).
(22)
ZHANG et al.: POWER SCALING OF UPLINK MASSIVE MIMO SYSTEMS WITH ARBITRARY-RANK CHANNEL MEANS
Similar to the perfect CSI case, the uplink sum rate per cell
can be defined as
969
(23)
where represents the coherence time of the channel, in terms
symbols are
of the number of symbols, during which
used as pilots for channel estimation.
which yields
user is
(31)
(24)
Proof: See Appendix I.
Note that the approximation in (24) does not require the
and
to be independent and becomes
random variables
more accurate as and increase. Thus, in massive MIMO
systems, due to the large number of BS antennas, this approximation will be particularly accurate.
where
Proof: Let
into (31), we obtain
.
, where
. Substituting it
A. Perfect CSI
We begin by considering the case with perfect CSI and establishing some key preliminary results which will be useful in
deriving our main results.
is asympLemma 2: By the law of large numbers, when
totically large, we have
(33)
We can also get
(25)
where -a.s. denotes almost sure convergence.
Proof: See Appendix II.
Lemma 3: The expectation for the inner product of two same
columns in can be found as
(34)
From Lemma 2, it is know that
(26)
and the expectation of the norm-square of the inner product of
any two columns in is given by
(35)
Then, (34) becomes5
(27)
(28)
where
(36)
Using
is defined as
(37)
(29)
5Note that the convergence of logarithms is sure, not almost sure as was
shown in [11].
970
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 8, NO. 5, OCTOBER 2014
(40)
where
Proof: Applying Lemma 1 on (31) yields
The three expectations in (41) can be evaluated using the re, the substitution of
sults in Lemma 3. Now, with
(26)(28) into (41) leads to the final result.
When
, (40) reduces to the special case of
Rayleigh fading channels:
(42)
It is known that
(43)
and the right hand side of (43) is the uplink rate lower bound
with perfect CSI in Rayleigh fading channels given by [12,
Proposition 2].
We use the expression in Theorem 2 to analyze how the uplink
rate changes with the Ricean -factor. The following corollary
presents the uplink rate limit when the Ricean -factor grows
without bound. To the best of our knowledge, this result is also
new.
, the approxCorollary 2: If for any and ,
imation in (40) converges to
(44)
6Note that
istic equivalent of
is not a constant value and is actually the determin. Therefore, it is a notational abuse to write
.
ZHANG et al.: POWER SCALING OF UPLINK MASSIVE MIMO SYSTEMS WITH ARBITRARY-RANK CHANNEL MEANS
.
, we can write (47) as
971
, i.e.,
(49)
which leads to
(50)
With the aid of Lemma 2, we can easily get the desired result.
It is interesting to note from Theorem 3 that with increasing
, the deterministic equivalent for the uplink rate of ZF
, is the same as that of MRC receivers,
.
receivers,
Hence, we can draw the conclusion that with no degradation
in the th users rate, in the case of ZF receivers and perfect
CSI, the transmit power of each user can be cut down at most
; then, the achievable uplink rate becomes
by
(51)
The following theorem gives a closed-form approximation
for the uplink rate using ZF receivers with perfect CSI. Note
that the steps to derive this expression are more complicated
than in the case with MRC receivers, since we need to deal with
complex non-central Wishart matrices.
Theorem 4: Using ZF receivers with perfect CSI, the achievable uplink rate of the th user can be approximated by
(52)
where
(53)
(58)
. Under the above assumpNow, let
tion, this is a Chi-squared random variable with distribution [34]
(59)
where
pressing the integral
(54)
Proof: Substituting
(63)
The entries of
if
if
.
.
(64)
7Note that this approximation has been often used in the context of MIMO
systems, e.g., [32], [33].
8The
Note that
denoted by
are known as
9Note
exists only if
[25].
972
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 8, NO. 5, OCTOBER 2014
When
(65)
becomes asymptotically an identity
Therefore,
matrix. Applying this result into (58) and then combining it
with (63), we obtain the desired result.
Corollary 5 shows that when
is large, similar to MRC, the
approximation for ZF also converges to the exact limit.
B. Imperfect CSI
In real scenarios, CSI is estimated at the receiver. A typical
channel estimation process using pilots is assumed in this paper
and has been described in Section II. Following a similar approach as for the case with perfect CSI, we first give some key
preliminary results.
Lemma 4: By the law of large numbers, when is asymptotically large, the inner product of any two columns in the channel
matrix
can be found as
(66)
(67)
where
, i.e.,
(68)
(69)
and the expectation for the norm-square of the inner product of
any two columns in is expressed as (70) (at the bottom of the
page).
Proof: See Appendix V.
1) MRC Receivers: For MRC, we have
. As a result,
from (22), the uplink rate of the th user with MRC receivers
can be written as (71) (at the bottom of the page).
Same as before, we will investigate the power-scaling law
but this time with imperfect CSI. We will first derive the exact
uplink rate and then an approximation for further analysis.
Theorem 5: Using MRC receivers with imperfect CSI from
MMSE estimation, if the transmit power of each user is scaled
, i.e.,
for
and a
down by a factor of
, we have
fixed
(72)
where
Proof: Let
, where
. Substituting it
into (71) yields (73) (at the bottom of the page). Let the numer, we obtain
ator and denominator of (73) be multiplied by
, using (67) in
(74) (at the bottom of the page). When
Lemma 4, it can be easily obtained that
(75)
if
if
(70)
(71)
(73)
(74)
ZHANG et al.: POWER SCALING OF UPLINK MASSIVE MIMO SYSTEMS WITH ARBITRARY-RANK CHANNEL MEANS
and with
(76)
we can further simplify (74) as
(77)
.
(78)
and
, we get
973
be cut down by
. This is reasonable since LOS propagation reduces fading fluctuations, thereby increasing the received
signal-to-interference-plus-noise ratio. As a consequence, with
no reduction in the rate performance, the transmit power for
non-zero Ricean -factor can be cut down more aggressively.
The following theorem presents a closed-form approximation
for the uplink rate using MRC with imperfect CSI.
Theorem 6: Using MRC receivers with imperfect CSI from
MMSE estimation, the achievable uplink rate of the th user
can be approximated by (82) (at the bottom of the page). where
.
Proof: Follow the same procedure as Theorem 2. After utilizing Lemma 1 and (69)(70), we can obtain the desired result
after some simplifications.
Note that when
,
reduces to the special
case of Rayleigh fading channel. After performing some simplifications, we have
(79)
Since
,
and we can further simplify (79) to
complete the proof.
It is important to note from Theorem 5 that the value of the deis dependent on both the Ricean
terministic equivalent for
-factor and the scaling parameter , which can be more pre, with increasing ,
cisely described as follows. When
only
can make
approach a non-zero constant
value. Otherwise, it will tend to zero with
and grow
, which will change the rate perwithout bound with
, we find that only
can
formance. For the case
lead
to a fixed value, and when
and
, it will
approximately become zero and , respectively. These observations can be summarized as the power-scaling law in the following corollary.
Corollary 6: With no degradation in the th users rate, using
MRC receivers with imperfect CSI and for a fixed , we obtain
the following power-scaling law. When the th users Ricean
-factor is zero, we can at most scale down the transmit power
and the achievable uplink rates becomes
to
(80)
On the other hand, with a non-zero Ricean
power can be at most scaled down to
(83)
It is known that
(84)
and the right hand side of (84) is the uplink rate lower bound
with imperfect CSI in Rayleigh fading channels given by [12,
Proposition 6].
Now, we consider the particular case when the Ricean
-factor grows infinite for the approximation in Theorem 6.
, the approxCorollary 7: If for any and ,
imation in (82) converges to
(81)
Corollary 6 reveals that with imperfect CSI, the amount of
.
power reduction achievable on the th user depends on
, the transmit power
For Rayleigh fading channels, i.e.,
, which agrees
can be at most cut down by a factor of
with the result in [12, Proposition 5]; however, for channels
, the transmit power can
with LOS components, i.e.,
(85)
(82)
974
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 8, NO. 5, OCTOBER 2014
Proof: Let
, where
. Then, (89) becomes (91) (at the bottom of the page). By Lemma 4, we know
that
(92)
where
(86)
for a fixed
, when
grows without bound,
If
(82) will converge to a non-zero constant value only when
and (82) becomes
(87)
Proof: With the fact that
(88)
the result then follows by performing some basic algebraic manipulations.
Again, it is noted that when
, the approximation
of the uplink rate becomes identical to the exact limit obtained
from Corollary 6.
.
2) ZF Receivers: For ZF receivers, we have
According to (22), the achievable uplink rate of ZF receivers
with imperfect CSI is given by
(93)
Due to
, as
into (93) yields
, the substitution of
(94)
and we can also easily get
(95)
(89)
Similar to the case with MRC receivers, we next investigate
the power-scaling law for ZF receivers.
Theorem 7: Using ZF receivers with imperfect CSI from
MMSE estimation, if the transmit power of each user is scaled
, i.e.,
for
and a
down by a factor of
, we have
fixed
(90)
where
(91)
ZHANG et al.: POWER SCALING OF UPLINK MASSIVE MIMO SYSTEMS WITH ARBITRARY-RANK CHANNEL MEANS
975
It is also important to point out that for both perfect and imperfect CSI, the exact limit of the uplink rate of ZF receivers
equals that of MRC receivers, which is consistent with the conclusion given in [12]. The following theorem presents a closedform approximation for the achievable uplink rate using ZF receivers with imperfect CSI.
Theorem 8: Using ZF receivers with imperfect CSI from
MMSE estimation, the achievable uplink rate of the th user
can be approximated by
where
(98)
(99)
is
a
diagonal
matrix
consisting
of
and
.
Proof: Applying Jensens inequality in (89), we get
(100)
The proof then follows the same procedure as in Theorem 4.
Corollary 9: If for any and ,
, the approximation in (98) converges to
(101)
Note that the fixed value in this case is the same as the limit
in Corollary 4 given by (61), which shows that for both MRC
and ZF receivers, as the Ricean -factor grows without bound,
the approximations of the uplink rate will tend to the same fixed
value regardless of the CSI quality.
for a fixed
, when
Corollary 10: If
grows without bound, the limit of (98) will exist only when
and (98) becomes
(102)
for a fixed
, when
grows without bound,
If
(98) will converge to a non-zero constant value only when
and (98) becomes
(103)
Proof: From the proof of Corollary 5, we know that when
is large,
becomes asymptotically an identity
matrix. With the fact that
(104)
the result then follows by some basic algebraic manipulations.
Fig. 1. Uplink sum rate per cell versus the number of BS antennas, with
users and transmit power
.
976
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 8, NO. 5, OCTOBER 2014
Fig. 2. Uplink sum rate per cell versus the number of BS antennas, with
users and scaled-down power
, where
.
Fig. 3. Uplink sum rate per cell versus the number of BS antennas with
users and scaled-down power
, where
.
Fig. 4. Uplink sum rate per cell versus the Ricean -factor with
users, number of BS antennas
and transmit power
ZHANG et al.: POWER SCALING OF UPLINK MASSIVE MIMO SYSTEMS WITH ARBITRARY-RANK CHANNEL MEANS
977
(110)
Comparing
(108)
and
(110),
we
find
that
is a quantity that lies between the upper and lower bound of
.
Therefore, it can be used as an approximation.
Moreover, we can quantify the offset between these two
bounds as
APPENDIX I
PROOF OF LEMMA 1
We know that
around
(111)
,
(105)
With the Jensens inequality, we can get the following bounds:
(112)
(106)
and
(107)
Combing (106) and (107), a lower and upper bound on (105)
can be obtained as follows
(108)
Utilizing
and
and combining them with the left and right hand side
of (108), respectively, we get that
(113)
Since both and are nonnegative, as and grow large,
and
will increase, and according to the law
and will approach their means, that
of large numbers,
is, their variances will become small. Hence, the offset decreases with the increase of and , that is, this approximation will be more and more accurate as the number of random
variables of and increase.
APPENDIX II
PROOF OF LEMMA 2
According to the law of large numbers, we know that
(114)
The entries of can be obtained from (4) and (5) described in
Section II-A. Hence,
(109)
(115)
978
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 8, NO. 5, OCTOBER 2014
with
and
representing the independent real and imag, respectively, each with zero-mean and
inary parts of
variance of 1/2. After substituting (115) into (114) and with the
definition
(116)
we have, as
(123)
(117)
If
(124)
To obtain the norm-square of (124), both the real and imaginary
parts should be extracted. Recalling (118), we can rewrite (124)
as
(118)
where
by using
the
is obtained
fact that
is bounded, we get as
(125)
(119)
Hence, we arrive at the desired result in (25).
APPENDIX III
PROOF OF LEMMA 3
Using (117) in Appendix II, we first consider the case
and we have
(126)
Then, with
(120)
(127)
substituting (125) and (126) into (127) and removing the terms
with zero expectation, we get the final result as
(128)
As we have obtained all expectations used in Lemma 3, we now
conclude the proof.
(122)
ZHANG et al.: POWER SCALING OF UPLINK MASSIVE MIMO SYSTEMS WITH ARBITRARY-RANK CHANNEL MEANS
APPENDIX IV
PROOF OF LEMMA 4
979
(135)
Now consider
(133), with
(129)
The first term which does not need to be estimated is the same as
, we use the MMSE estimain the case with perfect CSI. For
tion to get it and the details of the method have been introduced
in Section II-B2. Therefore, we have
(136)
we find (137) (at the bottom of the page). Since the entries of
and
are all i.i.d. zero-mean with unit-variance, many terms
in (137) have zero expectation. The only remaining term is the
element of the channel mean matrix. Hence, (137) can be written
as
(130)
where
,
and
have all been defined in (116) of
and
represent the independent real
Appendix III,
with zero-mean and variance 1/2,
and imaginary part of
is defined in (68). By the law of large numbers, it can be
and
got that
(138)
which can be simplified as
(131)
With (131) and the definition
(132)
we can obtain (133) (at the bottom of the page), as
.
Evaluating the expectation of all terms in (133) and removing
the terms with zero expectation, (133) can be simplified as
(139)
According to (119), we know that, as
(140)
As a consequence, (139) can be further simplified as
(134)
(141)
(133)
(137)
980
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 8, NO. 5, OCTOBER 2014
(143)
(144)
APPENDIX V
PROOF OF LEMMA 5
The expectation of
can be easily obtained by (134) in
Appendix IV. Hence, we have
(142)
Following the same procedure as in the proof of Lemma 3, we
can obtain the remaining three norm-square expectations. First
consider the case
. According to (133), we have (143)
(at the top of the page). After expanding the above equation
and removing all the terms with zero expectation, the remaining
terms are written as (144) (at the top of the page).
, we use the same method as in the proof
For the case
of Lemma 3. With (137), it is noted that the inner product
has non-zero real and imaginary parts. The result then follows
trivially by some algebraic manipulations.
REFERENCES
[1] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-time Wireless
Communications. Cambridge, U.K.: Cambridge Univ. Press, 2003.
[2] D. N. C. Tse and P. Viswanath, Fundamentals of Wireless Communication. Cambridge, U.K.: Cambridge Univ. Press, 2005.
[3] D. Gesbert, M. Kountouris, R. W. Heath Jr., C.-B. Chae, and T. Salzer,
Shifting the MIMO paradigm, IEEE Signal Process. Mag., vol. 24,
no. 5, pp. 3646, Sep. 2007.
[4] P. Viswanath and D. N. C. Tse, Sum capacity of the vector Gaussian
broadcast channel and uplink-downlink duality, IEEE Trans. Inf.
Theory, vol. 49, no. 8, pp. 19121921, Aug. 2003.
[5] G. Caire, N. Jindal, M. Kobayashi, and N. Ravindran, Multiuser
MIMO achievable rates with downlink training and channel state
feedback, IEEE Trans. Inf. Theory, vol. 56, no. 6, pp. 28452866,
Jun. 2010.
[6] J. Jose, A. Ashikhmin, T. L. Marzetta, and S. Vishwanath, Pilot contamination and precoding in multi-cell TDD systems, IEEE Trans.
Wireless Commun., vol. 10, no. 8, pp. 26402651, Aug. 2011.
[7] F. Rusek, D. Persson, B. K. Lau, E. G. Larsson, T. L. Marzetta, O.
Edfors, and F. Tufvesson, Scaling up MIMO: Opportunities and challenges with very large arrays, IEEE Signal Process. Mag., vol. 30, no.
1, pp. 4060, Jan. 2013.
[8] I. E. Telatar, Capacity of multi-antenna Gaussian channels, Europ.
Trans. Commun., pp. 585595, Nov.-Dec. 1999.
[9] T. L. Marzetta, How much training is required for multiuser
MIMO?, in Proc. Asilomar Conf. Signals, Syst., Comput., Oct. 2006,
pp. 359363.
[10] T. L. Marzetta, Noncooperative cellular wireless with unlimited numbers of base station antennas, IEEE Trans. Wireless Commun., vol. 9,
no. 11, pp. 35903600, Nov. 2010.
[11] J. Hoydis, S. ten Brink, and M. Debbah, Massive MIMO in the UL/DL
of cellular networks: How many antennas do we need?, IEEE J. Sel.
Areas Commun., vol. 31, no. 2, pp. 160171, Feb. 2013.
[12] H. Q. Ngo, E. G. Larsson, and T. L. Marzetta, Energy and spectral efficiency of very large multiuser MIMO systems, IEEE Trans. Commun.,
vol. 61, no. 4, pp. 14361449, Apr. 2013.
[13] A. Pitarokoilis, S. K. Mohammed, and E. G. Larsson, On the optimality of single-carrier transmission in large-scale antenna systems,
IEEE Wireless Commun. Lett., vol. 1, no. 4, pp. 276279, Aug. 2012.
[14] S. Wagner, R. Couillet, D. T. M. Slock, and M. Debbah, Large system
analysis of zero-forcing precoding in MISO broadcast channels with
limited feedback, in Proc. IEEE Int. Work. Signal Process. Adv. Wireless Commun. (SPAWC), Jun. 2010.
[15] A. M. Sayeed and N. Behdad, Continuous aperture phased MIMO: A
new architecture for optimum line-of-sight links, in Proc. IEEE Int.
Symp. Ant. Propag. (APS), Jul. 2011, pp. 293296.
[16] J. Brady, N. Behdad, and A. M. Sayeed, Beamspace MIMO for millimeter-wave communications: System architecture, modeling, analysis, and measurements, IEEE Trans. Antennas Propag., vol. 61, no.
7, pp. 38143827, Jul. 2013.
[17] A. M. Sayeed and J. Brady, Beamspace MIMO for high-dimensional
multiuser communication at millimeter-wave frequencies, in Proc.
IEEE Global Telecommun. Conf. (GLOBECOM), Dec. 2013.
[18] A. Lozano, A. M. Tulino, and S. Verd, Multiple-antenna capacity in
the low-power regime, IEEE Trans. Inf. Theory, vol. 49, no. 10, pp.
25272544, Oct. 2003.
[19] S. Jin, X. Q. Gao, and X. H. You, On the ergodic capacity of rank-1
Ricean fading MIMO channels, IEEE Trans. Inf. Theory, vol. 53, no.
2, pp. 502517, Feb. 2007.
[20] J. Zhang, C. K. Wen, S. Jin, X. Gao, and K.-K. Wong, On capacity
of large-scale MIMO multiple access channels with distributed sets of
correlated antennas, IEEE J. Sel. Areas Commun., vol. 31, no. 2, pp.
133148, Feb. 2013.
[21] C. K. Wen, S. Jin, and K.-K. Wong, On the sum-rate of multiuser
MIMO uplink channels with jointly-correlated Rician fading, IEEE
Trans. Commun., vol. 59, no. 10, pp. 28832895, Oct. 2011.
[22] C. K. Wen, K.-K. Wong, and J. C. Chen, Asymptotic mutual information for Rician MIMO-MA channels with arbitrary inputs: A replica
analysis, IEEE Trans. Commun., vol. 58, no. 10, pp. 27822788, Oct.
2010.
[23] J. G. Proakis, Digital Communications, 4th ed. New York, NY, USA:
McGraw-Hill, 2001.
[24] G. Alfano, A. Lozano, A. M. Tulino, and S. Verd, Mutual information
and eigenvalue distribution of MIMO Ricean channels, in Proc. IEEE
Int. Symp. Inf. Theory Appl. (ISITA), Oct. 2004, pp. 1013.
[25] M. Matthaiou, C. Zhong, and T. Ratnarajah, Novel generic bounds on
the sum rate of MIMO ZF receivers, IEEE Trans. Signal Process., vol.
59, no. 9, pp. 43414353, Sep. 2011.
[26] N. Ravindran, N. Jindal, and H. C. Huang, Beamforming with finite rate feedback for LOS MIMO downlink channels, in Proc. IEEE
Global Commun. Conf. (GLOBECOM), Nov. 2007, pp. 42004204.
[27] E. Torkildson, U. Madhow, and M. Rodwell, Indoor millimeter wave
MIMO: Feasibility and performance, IEEE Trans. Wireless Commun.,
vol. 10, no. 12, pp. 41504160, Dec. 2011.
ZHANG et al.: POWER SCALING OF UPLINK MASSIVE MIMO SYSTEMS WITH ARBITRARY-RANK CHANNEL MEANS
981