TSKS14
Multiple Antenna
Communications
Lecture 7, 2020
Emil Björnson
TSKS14 Multiple Antenna Communications 2020-04-21 2
Outline of this lecture
• Uplink Massive MIMO
• Pilot transmission
• Channel estimation
• Capacity lower bound
• Computation using channel estimates
• Maximization
TSKS14 Multiple Antenna Communications 2020-04-21 3
Recall: Uplink Massive MIMO system model
• Received signal:
𝒚 = 𝜌$% 𝑮𝒙 + 𝒘
where
𝑦+ 𝑔++ ⋯ 𝑔0+ 𝑥+ 𝑤+
𝒚= ⋮ 𝑮= ⋮ ⋱ ⋮ 𝒙= ⋮ 𝒘= ⋮
𝑦- 𝑔+- ⋯ 𝑔0- 𝑥0 𝑤-
• Parameters are normalized: Maximum power is 𝜌$%
𝑥+, … , 𝑥0 has power ≤ 1
• Channel of user 𝑘: 𝑔9+ , … , 𝑔9- ∼ 𝐶𝑁(0, 𝛽9 )
• Normalized noise: 𝒘 ∼ 𝐶𝑁(𝟎, 𝑰- )
Large-scale fading coefficient
TSKS14 Multiple Antenna Communications 2020-04-21 4
Sending pilot sequences
• Send pilot matrix 𝜏D 𝚽 = 𝜏D 𝝓+ … 𝝓0 over 𝜏D uses of the channel:
𝒀D = 𝜏D 𝜌$% 𝑮𝚽 H + 𝑾D
• Stacking received signals in 𝑀 × 𝜏D matrix 𝒀D 𝜏D 𝝓M
𝜏D 𝝓+
• Noise 𝑾D with i.i.d. 𝐶𝑁(0,1) elements
• Despreading of pilot signal:
𝒀LD = 𝒀D 𝚽 = 𝜏D 𝜌$% 𝑮𝚽 H 𝚽 + 𝑾D 𝚽
= 𝑰0 Still i.i.d. 𝐶𝑁(0,1)
elements
TSKS14 Multiple Antenna Communications 2020-04-21 5
Estimating Gaussian variable in noise
• Consider 𝑦 = 𝑝𝑔 + 𝑤 where
• 𝑝 is a constant, 𝑔 ∼ 𝐶𝑁 0, 𝛽 , 𝑤 ∼ 𝐶𝑁 0,1
Minimum mean squared error (MMSE) estimator:
Mean squared error:
𝑝𝛽
𝐸 𝑔P − 𝑔 M 𝑔P = 𝐸 𝑔|𝑦 = 𝑦
1 + 𝑝𝛽
𝑝𝛽M
Estimation error: 𝑔O = 𝑔P − 𝑔 ∼ 𝐶𝑁 0, 𝛽 − Independent
1 + 𝑝𝛽
𝑝𝛽M random
Estimate: 𝑔P ∼ 𝐶𝑁 0, variables
1 + 𝑝𝛽
TSKS14 Multiple Antenna Communications 2020-04-21 6
Estimates of the channels
𝒀LD T9
= 𝜏D 𝜌$% 𝑔9T + 𝑾D 𝚽 T9
• MMSE estimate of 𝑔9T from user 𝑘 to antenna 𝑚
• Estimate:
𝜏D 𝜌$% 𝛽9
T T L
𝑔P9 = 𝐸 𝑔9 |𝒀D = 𝒀LD T9 ∼ 𝐶𝑁(0, 𝛾9 )
1 + 𝜏D 𝜌$% 𝛽9
• Estimation error:
𝑔O9T = 𝑔P9T − 𝑔9T ∼ 𝐶𝑁 0, 𝛽9 − 𝛾9
where
𝜏D 𝜌$% 𝛽9M
𝛾9 =
1 + 𝜏D 𝜌$% 𝛽9
TSKS14 Multiple Antenna Communications 2020-04-21 7
How good is the channel estimate?
• Mean squared error (MSE):
M
𝜏 𝜌 𝛽
D $% 9
𝐸 𝑔P9T − 𝑔9T M = 𝐸 𝑔O9T M = 𝛽9 − 𝛾9 = 𝛽9 −
1 + 𝜏D 𝜌$% 𝛽9
• Goes to zero as 𝜌$% → ∞ or 𝜏D → ∞
(perfect estimate)
𝜏D = 10 𝜏D = 1
Simulation:
𝛽9 = 1 10% error
SNR = 𝜌$% 𝛽9 = 𝜌$%
TSKS14 Multiple Antenna Communications 2020-04-21 8
A capacity lower bound
Channel information Ω
𝑤
𝑥 𝜌 𝑔 + Receiver 𝑥P
𝑦 = 𝜌𝑔𝑥 + 𝑤
• Desired signal 𝑥, power 𝜌, and 𝑔 and 𝑤 uncorrelated
• Channel coefficient 𝑔, known channel information Ω
Capacity lower bound:
𝜌 𝐸 𝑔|Ω M
𝐶 ≥ 𝐸 log M 1 +
𝜌𝑉𝑎𝑟 𝑔|Ω + 𝑉𝑎𝑟 𝑤|Ω
TSKS14 Multiple Antenna Communications 2020-04-21 9
Uplink data transmission
• Received signal:
𝒚 = 𝜌$% 𝑮𝒙 + 𝒘
where
𝑦+ 𝑔++ ⋯ 𝑔0 + 𝑥+ 𝑤+
𝒚= ⋮ 𝑮= ⋮ ⋱ ⋮ 𝒙= ⋮ 𝒘= ⋮
𝑦- 𝑔+- ⋯ 𝑔0- 𝑥0 𝑤-
• Signals: 𝑥9 = 𝜂9 𝑞9 where 𝑞9 ∼ 𝐶𝑁 0,1 data symbol
0 ≤ 𝜂9 ≤ 1 controls the power
• Channel of user 𝑘: 𝑔9+ , … , 𝑔9- ∼ 𝐶𝑁(0, 𝛽9 )
• Normalized noise: 𝒘 ∼ 𝐶𝑁(𝟎, 𝑰- )
Large-scale fading coefficient
TSKS14 Multiple Antenna Communications 2020-04-21 10
Linear receiver processing
• Received signal:
+/M
𝒚 = 𝜌$% 𝑮𝑫𝜼 𝒒 + 𝒘
𝜂+ 0 0 𝑞+
where 𝑫𝜼 = 0 ⋱ 0 , 𝒒= ⋮
0 0 𝜂0 𝑞0
No successive
• Assign receiver filter 𝒂k for user 𝑖 interference cancelation
• Select it to make
+/M
𝒂Hk 𝒚 = 𝜌 𝒂H
$% k 𝑮𝑫 𝜼 𝒒 + 𝒂H
k 𝒘
0
= m 𝒂H 𝒈
k 9 𝜌 𝜂 𝑞
$% 9 9 + 𝒂H
k 𝒘 ≈ 𝑞k
9n+
In some “good” sense
TSKS14 Multiple Antenna Communications 2020-04-21 11
Capacity bound for User 𝑖 using Channel Estimates
• General capacity lower bound:
𝜌 𝐸 𝑔|Ω M
𝐶 ≥ 𝐸 log M 1+
𝜌𝑉𝑎𝑟 𝑔|Ω + 𝑉𝑎𝑟 𝑤′|Ω
In our case:
• 𝜌 = 𝜌$% 𝜂k
• 𝑔 = 𝒂H We need to compute each term!
k 𝒈k
• 𝑥 = 𝑞k
• Ω= 𝒈 r+, … , 𝒈
r0
• 𝑤 L = ∑0 𝒂H
9n+,9tk k 9𝒈 𝜌 𝜂 𝑞
$% 9 9 + 𝒂H
k 𝒘
TSKS14 Multiple Antenna Communications 2020-04-21 12
Computing the expectation in the numerator
𝐸 𝑔|Ω = 𝐸 𝒂H k 𝒈k |r r0
𝒈+, … , 𝒈
= 𝒂H
k 𝐸 r
𝒈 |r
𝒈
k + , … , r
𝒈 0 − 𝒂 H
u k |r
k 𝐸 𝒈 r0
𝒈+, … , 𝒈
= 𝒂H rk
k 𝒈
We have used
• 𝑔 = 𝒂H k 𝒈k
• 𝒂k selected based on Ω = 𝒈 r+, … , 𝒈
r0
• 𝒈k = 𝒈 rk − 𝒈
u k where 𝒈
r k and 𝒈
u k are independent
•𝐸 𝒈 u k |r r0 = 𝐸 𝒈
𝒈+, … , 𝒈 uk = 𝟎 𝜌 𝐸 𝑔|Ω M
𝐶 ≥ 𝐸 log M 1+
𝜌𝑉𝑎𝑟 𝑔|Ω + 𝑉𝑎𝑟 𝑤|Ω
TSKS14 Multiple Antenna Communications 2020-04-21 13
Computing the first term in the denominator
𝑉𝑎𝑟 𝑔|Ω = 𝐸 𝑔 M|Ω − 𝐸 𝑔|Ω M
• We have already computed 𝐸 𝑔|Ω
• Note that
M |Ω H M
𝐸 𝑔 = 𝐸 𝒂k 𝒈k |Ω
= 𝒂H
k 𝐸 𝒈r r
𝒈
k k
H
+ u
𝒈 u
𝒈
k k
H
− r
𝒈 u
𝒈
k k
H
− u
𝒈 r H
k k |Ω
𝒈 𝒂k
= 𝒂H
k r
𝒈 r H
k k + 𝛽k − 𝛾k 𝑰- − 𝟎 − 𝟎 𝒂k =
𝒈 𝒂H
k r
𝒈 r H
k k + 𝛽k − 𝛾k 𝑰- 𝒂k
𝒈
We used the same properties as before. 𝜌 𝐸 𝑔|Ω M
𝐶 ≥ 𝐸 log M 1+
𝜌𝑉𝑎𝑟 𝑔|Ω + 𝑉𝑎𝑟 𝑤|Ω
TSKS14 Multiple Antenna Communications 2020-04-21 14
Computing the second term in the denominator
0
𝑉𝑎𝑟 𝑤′|Ω = 𝑉𝑎𝑟 m 𝒂H 𝒈
k 9 𝜌 𝜂 𝑞
$% 9 9 + 𝒂H
k 𝒘|Ω
9n+,9tk
• 𝐸 𝑤′|Ω = 0 since 𝐸 𝑞9 = 0, 𝐸 𝒘 = 𝟎
𝑉𝑎𝑟 𝑤′|Ω = 𝐸 𝑤 L M|Ω
0
H M M |Ω H M
= m 𝐸 𝒂k 𝒈9 |Ω 𝜌$% 𝜂9 𝐸 𝑞9 +𝐸 𝒂k 𝒘 |Ω
9n+,9tk
0
= m 𝒂H rH
r9𝒈
k 𝒈
H
9 + 𝛽9 − 𝛾9 𝑰- 𝒂k 𝜌$% 𝜂9 + 𝒂k 𝑰- 𝒂k
9n+,9tk
TSKS14 Multiple Antenna Communications 2020-04-21 15
Result: Capacity lower bound of User 𝑖
H M
𝜌$% 𝜂k 𝒂k 𝒈rk
𝐶 ≥ 𝐸 log M 1+
𝒂Hk 𝑩k 𝒂k
where
0 0
𝑩k = m 𝜌$% 𝜂9 𝒈 rH
r9𝒈 9 + m 𝜌$% 𝜂9 𝛽9 − 𝛾9 𝑰- + 𝑰-
9n+,9tk 9n+
How to pick 𝒂k ?
TSKS14 Multiple Antenna Communications 2020-04-21 16
Generalized Rayleigh Quotient
The ratio
𝒂H 𝒃 M
𝒂H 𝑩𝒂
for given 𝒃 and invertible 𝑩 is maximized by 𝒂 = 𝑩z+𝒃
• Intuition:
y
𝒂w 𝒃
• The ratio 𝒂w 𝒂
is maximized by 𝒂 = 𝒃 Extension of maximum
ratio combining
• The extra term 𝑩z+ is “whitening”
TSKS14 Multiple Antenna Communications 2020-04-21 17
Maximizing the capacity lower bound
H M
𝒂k 𝒃k
𝐶 ≥ 𝐸 log M 1 +
𝒂H
k 𝑩k 𝒂k
r k and
where 𝒃k = 𝜌$% 𝜂k 𝒈
0 0
𝑩k = m 𝜌$% 𝜂9 𝒈 rH
r9𝒈 9 + m 𝜌$% 𝜂9 𝛽9 − 𝛾9 𝑰- + 𝑰-
9n+,9tk 9n+
Maximized by selecting
𝒂k = 𝑩z+
k 𝒃 k = 𝜌$% 𝜂 k 𝑩 z+
k 𝒈rk
Called MMSE combining
TSKS14 Multiple Antenna Communications 2020-04-21 18
z+
Interpretation of 𝒂k = 𝜌$% 𝜂k 𝑩k 𝒈rk (whitening)
rk
𝒈 𝒂k
Take 𝒈 r k and
rotate it using 𝑩z+
k
r+, 𝒈
𝒈 r kz+, 𝒈
r k{+, 𝒈
r0
TSKS14 Multiple Antenna Communications 2020-04-21 19
Summary
• Channel estimation
• Send pilot sequences of length 𝜏D ≥ 𝐾 in the uplink
• Estimate channels using MMSE estimation
• Data transmission
• General capacity lower bound
• Can be computed when MMSE channel estimates are known
• The linear receiver was optimized
End of Lecture 7
TSKS14 Multiple Antenna
Communications