Power Control in Multi-Cell Massive MIMO Systems

Dr. Salil Kashyap

Dept. of Electronics and Electrical Engineering
IIT Guwahati, India

May 7, 2020

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 1 / 20
Recap: Why is power control required?

• To handle near-far effects

⋄ Strong signal from a user nearer to the BS may mask the weak signal
from a user that is far-off, if power control is not done
• Enables uniformly good service throughout the cell
• Occurs on a long-time scale, since effective SINR depends only on the
large-scale fading coefficients that vary at a much slower pace

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 2 / 20
Generic Form of Effective SINR for Multi-Cell Systems

• Effective SINR for user k in cell l can be expressed in a generic form as

alk ηlk
SINRlk = K K
, (1)
X X
l ′k ′
XX ′ ′ X ′
1+ blk ηl ′ k ′ + clkl k ηl ′ k ′ + dlkl ηl ′ k
l ′ ∈Pl k ′ =1 / l k ′ =1
l ′ ∈P l ′ ∈Pl \{l}

′ ′ ′ ′ ′
l k
where {alk }, {blk }, {clkl k }, and {dlkl } are non-negative coefficients as given
in Table 1.

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 3 / 20
Summary of Coefficients for a Multi-Cell System

Table 1: Expressions for coefficients for a multi-cell system

Zero-Forcing Maximum-Ratio
Uplink alk = (M − K )ρul γlkl alk = Mρul γlkl
l ′k ′ l ′k ′
blk = ρul (βll′ k ′ − γll′ k ′ ) blk = ρul βll′ k ′
′ ′ ′ ′
clkl k = ρul βll′ k ′ clkl k = ρul βll′ k ′
′ ′
dlkl = (M − K )ρul γll′ k dlkl = Mρul γll′ k
Downlink alk = (M − K )ρdl γlkl alk = Mρdl γlkl
l ′k ′ ′ ′
l ′k ′ ′
blk = ρdl (βlkl − γlkl ) blk = ρdl βlkl
′ ′ ′ ′ ′ ′
clkl k = ρdl βlkl clkl k = ρdl βlkl
′ ′ ′ ′
dlkl = (M − K )ρdl γlkl dlkl = Mρdl γlkl

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 4 / 20
Power Control Coefficients for a Multi-Cell System

• The numerator and the denominator in the SINR expression in (1) is linear in
the power control coefficients {ηlk }
• The power control coefficient ηlk satisfy the conditions as specified in the
table below.

Table 2: Summary of constraints on power control coefficients for a multi-cell system

Multi-Cell
Uplink 0 ≤ ηlk ≤ 1
k = 1, . . . , K , l = 1, . . . , L
PK
Downlink k=1 ηlk = 1, l = 1, . . . , L
and
ηlk ≥ 0, k = 1, . . . , K , l = 1, . . . , L

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 5 / 20
Power Control With Given SINR Targets

• We next show how the problem of designing power control policy that offers
guaranteed quality-of-service can be cast as a linear feasibility problem
• Consider a set of constraints of the form

SINRlk ≥ SINRlk , k = 1, . . . , K , l = 1, . . . , L (2)

where SINRlk is a target SINR for the k th user in the l th cell.

• Note that an SINR target translates to a spectral efficiency target (QoS
requirement)
• Based on (1), the constraint in (2) is equivalent to the following set of
inequalities
 
XX K K
XX X
′ ′ ′ ′ ′
l k
alk ηlk ≥ SINRlk 1 + blk ηl ′ k ′ + clkl k ηl ′ k ′ + dlkl ηl ′ k  ,
l ′ ∈Pl k ′ =1 l ′ ∈P
/ l k ′ =1 l ′ ∈Pl \{l}
(3)
⋄ The set of inequalities above are linear in {ηlk }

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 6 / 20
Problem of Power Control Policy Design

• The problem of designing power control policy under which the k th user in
the l th cell achieves an SINR of at least SINRlk can be stated mathematically
as follows:

find {ηlk }
subject to (i) SINRlk ≥ SINRlk , k = 1, . . . , K , l = 1, . . . , L,
(ii) the constraints in Table 2. (4)

• Note that (4) is a linear programming feasibility problem

⋄ Can be solved using standard software toolboxes

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 7 / 20
Max-Min Fairness Power Control

• For single-cell system, max-min fairness means that SINR target of all users
in the cell are equal
• In a multi-cell system, max-min fairness may be imposed network-wide or
independently within each cell

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 8 / 20
Multi-Cell Systems with Network-Wide Max-Min Fairness

• Under network-wide max-min fairness, all target SINRs are set equal, i.e.,

SINRlk = SINR, k = 1, . . . , K , l = 1, . . . , L. (5)

• Max-min philosophy results in the following optimization problem

maximize SINR
with respect to {ηlk }
subject to (i) SINRlk ≥ SINR, k = 1, . . . , K , l = 1, . . . , L,
(ii) the constraints in Table 2. (6)

• All inequalities in (6) are linear

• Note that (6) gives power control coefficients such that users in all cells
achieve the same SINR
• The power control coefficients in cell l depend on the conditions in other
cells, l ′ 6= l that are far-away

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 9 / 20
Problem with Imposing Network-Wide Max-Min Fairness

• Under network-wide max-min fairness, all target SINRs are set equal, i.e.,

SINRlk = SINR, k = 1, . . . , K , l = 1, . . . , L. (7)

• If some cell in the network has low throughput

⋄ due to overcrowding with users
⋄ due to the user scheduled for service experiencing severe shadow fading
(bad channel)
• Then this low throughput is unnecessarily imposed on all users served in all
other cells
• The value of SINR achieved in (6) approaches 0 as the number of cells
L→∞
• Therefore, network-wide max-min fairness is not scalable with respect to the
number of cells

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 10 / 20
Per Cell Power Control

• An approach widely adopted to solve the scalability of network-wide max-min

fairness:
• Equalize the SINRs only within each cell
• Will discuss per cell power control under two assumptions:
′
• A1: Negligible coherent interference, i.e., dlkl = 0 for all l, l ′ and k. Note
that this corresponds to excluding the fourth term in the denominator
of (1) that arises due to pilot contamination.
• A2: Each cell uses the full available power
◦ On the UL, ηlk = 1, for some k for every l = 1, . . . , L, i.e., at least
one user in each cell transmits with full power
◦ On
PKthe DL, all BSs spend the maximum available power, i.e.,
k=1 ηlk = 1, l = 1, . . . , L

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 11 / 20
Features of Per Cell Power Control

• Under A1 and A2, max-min can be performed in each cell independently

• At the optimum, all users in each cell achieve a cell-specific target SINR
(SINRl )
• Under per-cell power control,
⋄ Each cell is equally important
⋄ No cell dictates what other cells should do
• Under network-wide power control,
⋄ The achievable throughput is dictated by the most disadvantaged cell

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 12 / 20
Per Cell Uplink Power Control

• Under the assumption A1, effective SINR in (1) simplifies to

alk ηlk
SINRlk = K K
, (8)
X X XX
l ′k ′ ′ ′
1+ blk ηl ′ k ′ + clkl k ηl ′ k ′
l ′ ∈Pl k ′ =1 l ′ ∈P
/ l k ′ =1

′ ′ ′ ′
l k
• On the UL, {blk } and {clkl k } are independent of k as can be seen from
Table 1.
′ ′
l ′k ′ ′ ′ ′ ′
• Let us denote bll k = blk and cll k = clkl k , then (8) further simplifies to
alk ηlk
SINRlk = K K
, (9)
XX ′ ′ XX ′ ′
1+ bll k ηl ′ k ′ + cll k ηl ′ k ′
l ′ ∈Pl k ′ =1 l ′ ∈P
/ l k ′ =1

• In (9), denominator is independent of k, and SINRlk has the same form as in

single-cell case

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 13 / 20
Per Cell Uplink Power Control Contd.
• Due to the assumption A2, the technique for single-cell can be applied
• The following choice gives max-min fairness in the l th cell, for all l
min{alk ′ }
k
′
ηlk = , k = 1, . . . , K , l = 1, . . . , L (10)
alk
• And the SINR achieved by all users in the l th cell equals
min{alk ′ }
′k
SINRl = K K
,
XX ′ ′ XX ′ ′
1+ bll k ηl ′ k ′ + cll k ηl ′ k ′
l ′ ∈Pl k ′ =1 l ′ ∈P
/ l k ′ =1
1
=
X min
k′
{a l ′k ′ } X
K ′ ′
bl k X min
k′
{al ′ k ′ } X
K ′ ′
cl k
1 l l
+ +
min {alk ′ } min {alk ′ } al ′ k ′ min {alk ′ } al ′ k ′
′k l ′ ∈Pl k
′ k ′ =1 l ′ ∈P
/ l k
′ k ′ =1
(11)
• This is independent of k, insert formulas for coefficients from Table 1 to
obtain results for different scenarios
(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 14 / 20
Per Cell Downlink Power Control

• Under the assumption A1, effective SINR in (1) simplifies to

alk ηlk
SINRlk = K K
, (12)
X X
l ′k ′
XX ′ ′
1+ blk ηl ′ k ′ + clkl k ηl ′ k ′
l ′ ∈Pl k ′ =1 l ′ ∈P
/ l k ′ =1

′ ′ ′ ′
l k
• On the DL, {blk } and {clkl k } are independent of k ′ as can be seen from
Table 1.
′ ′ ′ ′ ′ ′
l l k
• Let us denote blk = blk and clkl = clkl k , and under the assumption A2 for
DL, (12) further simplifies to
alk ηlk
SINRlk = X ′ X ′, (13)
l
1+ blk + clkl
l ′ ∈Pl l ′ ∈P
/ l

• In (13), denominator is independent of k ′ , and SINRlk has the same form as

in single-cell case

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 15 / 20
Per Cell Downlink Power Control Contd.
• Using similar argument as for the single-cell DL, the power control
coefficients that yield max-min optimality in each cell equal
X ′ X ′
l
1+ blk + clkl
l ′ ∈Pl l ′ ∈P
/ l
ηlk = X X (14)
l′ ′
1+ blk + clkl
K
X l ′ ∈Pl l ′ ∈P
/ l
alk
a lk ′′
k ′′ =1

• Resulting max-min SINR achieved by all users in cell l equals

1
SINRl = X ′ X ′ (15)
l
1+ blk + clkl
K
X l ′ ∈Pl l ′ ∈P
/ l
alk ′′
k ′′ =1

• Note that this is independent of k

• Insert formulas for coefficients from Table 1 to obtain results for different DL
scenarios
(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 16 / 20
Discussion on Max-Min

• Max-Min fairness power control ensures

⋄ Uniform quality of service for all users
• Due to path loss and shadowing, some users are likely to have a very small
βll′ k
• In max-min power control, the worst user is usually the bottleneck and drives
network performance
• Some of the worst users may be dropped from service before computing the
max-min coefficients in order to improve the throughput of the network.

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 17 / 20
What next?

Last Topic: Propagation Channel

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 18 / 20
References:

• T. L. Marzetta, E. G. Larsson, H. Yang, H. Q. Ngo, “Fundamentals of

Massive MIMO,” Cambridge University Press, 2016

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 19 / 20
 Thank You!

(Salil Kashyap, EEE, IITG) EE 638: Power Control: Multiple Cell May 7, 2020 20 / 20