Effect of antenna correlation on Massive MIMO (2)

Rohit Budhiraja

MIMO Wireless Communications (EE677)

Nov 5, 2019

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 1

 Impact of spatial correlation on eigenvalues of R
Eigen structure of R determines the spatial channel correlation of the channel h

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 2

 Impact of spatial correlation on eigenvalues of R
Eigen structure of R determines the spatial channel correlation of the channel h
which spatial directions are statistically more likely to contain strong signal components than others

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 2

 Impact of spatial correlation on eigenvalues of R
Eigen structure of R determines the spatial channel correlation of the channel h
which spatial directions are statistically more likely to contain strong signal components than others
strong spatial correlation is characterized by large eigenvalue variations.

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 2

 Impact of spatial correlation on eigenvalues of R
Eigen structure of R determines the spatial channel correlation of the channel h
which spatial directions are statistically more likely to contain strong signal components than others
strong spatial correlation is characterized by large eigenvalue variations.

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 2

 Impact of spatial correlation on eigenvalues of R
Eigen structure of R determines the spatial channel correlation of the channel h
which spatial directions are statistically more likely to contain strong signal components than others
strong spatial correlation is characterized by large eigenvalue variations.

For uncorrelated fading R = IM . Correlation matrices are normalized such that Tr(R) = M

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 2

 Impact of spatial correlation on eigenvalues of R
Eigen structure of R determines the spatial channel correlation of the channel h
which spatial directions are statistically more likely to contain strong signal components than others
strong spatial correlation is characterized by large eigenvalue variations.

For uncorrelated fading R = IM . Correlation matrices are normalized such that Tr(R) = M
Spatial channel correlation makes 30 of the 100 eigenvalues larger than in uncorrelated case

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 2

 Impact of spatial correlation on eigenvalues of R
Eigen structure of R determines the spatial channel correlation of the channel h
which spatial directions are statistically more likely to contain strong signal components than others
strong spatial correlation is characterized by large eigenvalue variations.

For uncorrelated fading R = IM . Correlation matrices are normalized such that Tr(R) = M
Spatial channel correlation makes 30 of the 100 eigenvalues larger than in uncorrelated case
Spatial channel correlation reduces the remaining eigenvalues to substantially smaller values

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 2

 Impact of spatial correlation on eigenvalues of R
Eigen structure of R determines the spatial channel correlation of the channel h
which spatial directions are statistically more likely to contain strong signal components than others
strong spatial correlation is characterized by large eigenvalue variations.

For uncorrelated fading R = IM . Correlation matrices are normalized such that Tr(R) = M
Spatial channel correlation makes 30 of the 100 eigenvalues larger than in uncorrelated case
Spatial channel correlation reduces the remaining eigenvalues to substantially smaller values
We can generate any random channel vector h ∼ NC (0M , R)

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 2

 Impact of spatial correlation on eigenvalues of R
Eigen structure of R determines the spatial channel correlation of the channel h
which spatial directions are statistically more likely to contain strong signal components than others
strong spatial correlation is characterized by large eigenvalue variations.

For uncorrelated fading R = IM . Correlation matrices are normalized such that Tr(R) = M
Spatial channel correlation makes 30 of the 100 eigenvalues larger than in uncorrelated case
Spatial channel correlation reduces the remaining eigenvalues to substantially smaller values
We can generate any random channel vector h ∼ NC (0M , R)
1 1
perfom eigen value decomposition of R = UDUH ; take its square root R 2 = UD 2 UH

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 2

 Impact of spatial correlation on eigenvalues of R
Eigen structure of R determines the spatial channel correlation of the channel h
which spatial directions are statistically more likely to contain strong signal components than others
strong spatial correlation is characterized by large eigenvalue variations.

For uncorrelated fading R = IM . Correlation matrices are normalized such that Tr(R) = M
Spatial channel correlation makes 30 of the 100 eigenvalues larger than in uncorrelated case
Spatial channel correlation reduces the remaining eigenvalues to substantially smaller values
We can generate any random channel vector h ∼ NC (0M , R)
1 1
perfom eigen value decomposition of R = UDUH ; take its square root R 2 = UD 2 UH
1
generate ě ∼ NC (0M , IM ) and perform h = R 2 ě

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 2

 Impact of spatial correlation on eigenvalues of R
Eigen structure of R determines the spatial channel correlation of the channel h
which spatial directions are statistically more likely to contain strong signal components than others
strong spatial correlation is characterized by large eigenvalue variations.

For uncorrelated fading R = IM . Correlation matrices are normalized such that Tr(R) = M
Spatial channel correlation makes 30 of the 100 eigenvalues larger than in uncorrelated case
Spatial channel correlation reduces the remaining eigenvalues to substantially smaller values
We can generate any random channel vector h ∼ NC (0M , R)
1 1
perfom eigen value decomposition of R = UDUH ; take its square root R 2 = UD 2 UH
1
generate ě ∼ NC (0M , IM ) and perform h = R 2 ě
H
note that E{hh } = R
MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 2
 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
Rk = K Uk UH
k

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

here Uk ∈ CM×M/K are tall unitary matrices (UHk Uk = IM/K )

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

here Uk ∈ CM×M/K are tall unitary matrices (UHk Uk = IM/K ) and UHk Uj = 0M , ∀k 6= j

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

here Uk ∈ CM×M/K are tall unitary matrices (UHk Uk = IM/K ) and UHk Uj = 0M , ∀k 6= j
factor K normalizes the average channel gain such that βk = M1 Tr(Rk ) = 1

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

here Uk ∈ CM×M/K are tall unitary matrices (UHk Uk = IM/K ) and UHk Uj = 0M , ∀k 6= j
factor K normalizes the average channel gain such that βk = M1 Tr(Rk ) = 1
note that Eke k k2 = M/K (will later use this fact)

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

here Uk ∈ CM×M/K are tall unitary matrices (UHk Uk = IM/K ) and UHk Uj = 0M , ∀k 6= j
factor K normalizes the average channel gain such that βk = M1 Tr(Rk ) = 1
note that Eke k k2 = M/K (will later use this fact)
Channel model implies that each UE has a strongly spatially correlated channel

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

here Uk ∈ CM×M/K are tall unitary matrices (UHk Uk = IM/K ) and UHk Uj = 0M , ∀k 6= j
factor K normalizes the average channel gain such that βk = M1 Tr(Rk ) = 1
note that Eke k k2 = M/K (will later use this fact)
Channel model implies that each UE has a strongly spatially correlated channel
with only M/K rather than M non-zero eigenvalues of R

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

here Uk ∈ CM×M/K are tall unitary matrices (UHk Uk = IM/K ) and UHk Uj = 0M , ∀k 6= j
factor K normalizes the average channel gain such that βk = M1 Tr(Rk ) = 1
note that Eke k k2 = M/K (will later use this fact)
Channel model implies that each UE has a strongly spatially correlated channel
with only M/K rather than M non-zero eigenvalues of R
Eigenspaces of individual correlation matrices are all orthogonal.

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

here Uk ∈ CM×M/K are tall unitary matrices (UHk Uk = IM/K ) and UHk Uj = 0M , ∀k 6= j
factor K normalizes the average channel gain such that βk = M1 Tr(Rk ) = 1
note that Eke k k2 = M/K (will later use this fact)
Channel model implies that each UE has a strongly spatially correlated channel
with only M/K rather than M non-zero eigenvalues of R
Eigenspaces of individual correlation matrices are all orthogonal.
implies although the UEs’ channels are random, they “live” in mutually orthogonal subspaces

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

here Uk ∈ CM×M/K are tall unitary matrices (UHk Uk = IM/K ) and UHk Uj = 0M , ∀k 6= j
factor K normalizes the average channel gain such that βk = M1 Tr(Rk ) = 1
note that Eke k k2 = M/K (will later use this fact)
Channel model implies that each UE has a strongly spatially correlated channel
with only M/K rather than M non-zero eigenvalues of R
Eigenspaces of individual correlation matrices are all orthogonal.
implies although the UEs’ channels are random, they “live” in mutually orthogonal subspaces
K
X
y = hi si + n
i =1

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

here Uk ∈ CM×M/K are tall unitary matrices (UHk Uk = IM/K ) and UHk Uj = 0M , ∀k 6= j
factor K normalizes the average channel gain such that βk = M1 Tr(Rk ) = 1
note that Eke k k2 = M/K (will later use this fact)
Channel model implies that each UE has a strongly spatially correlated channel
with only M/K rather than M non-zero eigenvalues of R
Eigenspaces of individual correlation matrices are all orthogonal.
implies although the UEs’ channels are random, they “live” in mutually orthogonal subspaces
K
X
y = hi si + n
i =1
K
!
X
UHk y = UHk hi si + n
i =1

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

here Uk ∈ CM×M/K are tall unitary matrices (UHk Uk = IM/K ) and UHk Uj = 0M , ∀k 6= j
factor K normalizes the average channel gain such that βk = M1 Tr(Rk ) = 1
note that Eke k k2 = M/K (will later use this fact)
Channel model implies that each UE has a strongly spatially correlated channel
with only M/K rather than M non-zero eigenvalues of R
Eigenspaces of individual correlation matrices are all orthogonal.
implies although the UEs’ channels are random, they “live” in mutually orthogonal subspaces
K
X
y = hi si + n
i =1
K
! K
X X √
UHk y = UHk hi si + n = K UHk Ui e i si + UHk n
i =1 i =1

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (1)
Consider uplink of a single-cell with K users. Assume hk ∼ (0M , Rk ), k = 1, . . . , K
We make (slightly artificial) assumption that
√
Rk = K Uk UH


k ⇒ hk = K Uk e k , where e k ∼ NC 0M/K , IM/K

here Uk ∈ CM×M/K are tall unitary matrices (UHk Uk = IM/K ) and UHk Uj = 0M , ∀k 6= j
factor K normalizes the average channel gain such that βk = M1 Tr(Rk ) = 1
note that Eke k k2 = M/K (will later use this fact)
Channel model implies that each UE has a strongly spatially correlated channel
with only M/K rather than M non-zero eigenvalues of R
Eigenspaces of individual correlation matrices are all orthogonal.
implies although the UEs’ channels are random, they “live” in mutually orthogonal subspaces
K
X
y = hi si + n
i =1
K
! K
X X √ √
UHk y = UHk hi si + n = K UHk Ui e i si + UHk n = K e k sk + ň k
i =1 i =1

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 3

 Basic impact of spatial channel correlation (2)
√
We have UH
ky = K e k sk + ňk

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Basic impact of spatial channel correlation (2)
√
We have UH ky = K e k sk + ňk
Due to structure of spatial correlation matrices, multiuser channel

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Basic impact of spatial channel correlation (2)
√
We have UH ky = K e k sk + ňk
Due to structure of spatial correlation matrices, multiuser channel
is divided into K orthogonal single-user channels with no interference

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Basic impact of spatial channel correlation (2)
√
We have UH ky = K e k sk + ňk
Due to structure of spatial correlation matrices, multiuser channel
is divided into K orthogonal single-user channels with no interference
Average SINR of UE k

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Basic impact of spatial channel correlation (2)
√
We have UH ky = K e k sk + ňk
Due to structure of spatial correlation matrices, multiuser channel
is divided into K orthogonal single-user channels with no interference
Average SINR of UE k

Kpk ke k k2
 
E{SINRk } = E 2
σUL

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Basic impact of spatial channel correlation (2)
√
We have UH ky = K e k sk + ňk
Due to structure of spatial correlation matrices, multiuser channel
is divided into K orthogonal single-user channels with no interference
Average SINR of UE k

Kpk ke k k2 Kpk Eke k k2

 
M pk
E{SINRk } = E 2 = 2 = 2
σUL σUL σUL

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Basic impact of spatial channel correlation (2)
√
We have UH ky = K e k sk + ňk
Due to structure of spatial correlation matrices, multiuser channel
is divided into K orthogonal single-user channels with no interference
Average SINR of UE k

Kpk ke k k2 Kpk Eke k k2

 
M pk
E{SINRk } = E 2 = 2 = 2
σUL σUL σUL

Signal power of each UE gets a boost of M – antenna array captures the same amount of energy

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Basic impact of spatial channel correlation (2)
√
We have UH ky = K e k sk + ňk
Due to structure of spatial correlation matrices, multiuser channel
is divided into K orthogonal single-user channels with no interference
Average SINR of UE k

Kpk ke k k2 Kpk Eke k k2

 
M pk
E{SINRk } = E 2 = 2 = 2
σUL σUL σUL

Signal power of each UE gets a boost of M – antenna array captures the same amount of energy
but this energy is concentrated on a subset of the eigen/spatial directions

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Basic impact of spatial channel correlation (2)
√
We have UH ky = K e k sk + ňk
Due to structure of spatial correlation matrices, multiuser channel
is divided into K orthogonal single-user channels with no interference
Average SINR of UE k

Kpk ke k k2 Kpk Eke k k2

 
M pk
E{SINRk } = E 2 = 2 = 2
σUL σUL σUL

Signal power of each UE gets a boost of M – antenna array captures the same amount of energy
but this energy is concentrated on a subset of the eigen/spatial directions
We compare the above case with the one where all UEs have same R

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Basic impact of spatial channel correlation (2)
√
We have UH ky = K e k sk + ňk
Due to structure of spatial correlation matrices, multiuser channel
is divided into K orthogonal single-user channels with no interference
Average SINR of UE k

Kpk ke k k2 Kpk Eke k k2

 
M pk
E{SINRk } = E 2 = 2 = 2
σUL σUL σUL

Signal power of each UE gets a boost of M – antenna array captures the same amount of energy
but this energy is concentrated on a subset of the eigen/spatial directions
We compare the above case with the one where all UEs have same R

R = K UUH

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Basic impact of spatial channel correlation (2)
√
We have UH ky = K e k sk + ňk
Due to structure of spatial correlation matrices, multiuser channel
is divided into K orthogonal single-user channels with no interference
Average SINR of UE k

Kpk ke k k2 Kpk Eke k k2

 
M pk
E{SINRk } = E 2 = 2 = 2
σUL σUL σUL

Signal power of each UE gets a boost of M – antenna array captures the same amount of energy
but this energy is concentrated on a subset of the eigen/spatial directions
We compare the above case with the one where all UEs have same R
√
R = K UUH ⇒ hk = K Ue k ,

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Basic impact of spatial channel correlation (2)
√
We have UH ky = K e k sk + ňk
Due to structure of spatial correlation matrices, multiuser channel
is divided into K orthogonal single-user channels with no interference
Average SINR of UE k

Kpk ke k k2 Kpk Eke k k2

 
M pk
E{SINRk } = E 2 = 2 = 2
σUL σUL σUL

Signal power of each UE gets a boost of M – antenna array captures the same amount of energy
but this energy is concentrated on a subset of the eigen/spatial directions
We compare the above case with the one where all UEs have same R
√
R = K UUH ⇒ hk = K Ue k , where e k ∼ NC 0M/K , IM/K

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Basic impact of spatial channel correlation (2)
√
We have UH ky = K e k sk + ňk
Due to structure of spatial correlation matrices, multiuser channel
is divided into K orthogonal single-user channels with no interference
Average SINR of UE k

Kpk ke k k2 Kpk Eke k k2

 
M pk
E{SINRk } = E 2 = 2 = 2
σUL σUL σUL

Signal power of each UE gets a boost of M – antenna array captures the same amount of energy
but this energy is concentrated on a subset of the eigen/spatial directions
We compare the above case with the one where all UEs have same R
√
R = K UUH ⇒ hk = K Ue k , where e k ∼ NC 0M/K , IM/K

Here U ∈ CM×M/K are tall unitary matrices (UH U = IM/K )

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 4

 Summary of impact of spatial channel correlation

We have
K
X
y = h i si + n
i=1

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 5

 Summary of impact of spatial channel correlation

We have
K
X
y = h i si + n
i=1
K
!
X
H H
U y = U h i si + n
i=1

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 5

 Summary of impact of spatial channel correlation

We have
K
X
y = h i si + n
i=1
K
! K √
X X
H H
U y = U h i si + n = K UH Ue i si + UH
kn
i=1 i=1

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 5

 Summary of impact of spatial channel correlation

We have
K
X
y = h i si + n
i=1
K
! K √ K
H H
X X √ X
U y = U h i si + n = K UH Ue i si + UH
k n = K e i si + ň
i=1 i=1 i=1

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 5

 Summary of impact of spatial channel correlation

We have
K
X
y = h i si + n
i=1
K
! K √ K
H H
X X √ X
U y = U h i si + n = K UH Ue i si + UH
k n = K e i si + ň
i=1 i=1 i=1

This is not a single-user channel, but a K-user channel

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 5

 Summary of impact of spatial channel correlation

We have
K
X
y = h i si + n
i=1
K
! K √ K
H H
X X √ X
U y = U h i si + n = K UH Ue i si + UH
k n = K e i si + ň
i=1 i=1 i=1

This is not a single-user channel, but a K-user channel

Individual spatial correlation matrices do not affect the system behavior

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 5

 Summary of impact of spatial channel correlation

We have
K
X
y = h i si + n
i=1
K
! K √ K
H H
X X √ X
U y = U h i si + n = K UH Ue i si + UH
k n = K e i si + ň
i=1 i=1 i=1

This is not a single-user channel, but a K-user channel

Individual spatial correlation matrices do not affect the system behavior but the collection of all
UEs’ correlation matrices affect

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 5

 Summary of impact of spatial channel correlation

We have
K
X
y = h i si + n
i=1
K
! K √ K
H H
X X √ X
U y = U h i si + n = K UH Ue i si + UH
k n = K e i si + ň
i=1 i=1 i=1

This is not a single-user channel, but a K-user channel

Individual spatial correlation matrices do not affect the system behavior but the collection of all
UEs’ correlation matrices affect
Spatial channel correlation can be very beneficial in massive MIMO if

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 5

 Summary of impact of spatial channel correlation

We have
K
X
y = h i si + n
i=1
K
! K √ K
H H
X X √ X
U y = U h i si + n = K UH Ue i si + UH
k n = K e i si + ň
i=1 i=1 i=1

This is not a single-user channel, but a K-user channel

Individual spatial correlation matrices do not affect the system behavior but the collection of all
UEs’ correlation matrices affect
Spatial channel correlation can be very beneficial in massive MIMO if
UEs have sufficiently different spatial correlation matrices

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 5

 Local Scattering Spatial Correlation Model (1)
Develop a model for R ∈ CM×M for a NLoS channel between a UE and a BS equipped with ULA
UE and BS indices are dropped for simplicity

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 6

 Local Scattering Spatial Correlation Model (1)
Develop a model for R ∈ CM×M for a NLoS channel between a UE and a BS equipped with ULA
UE and BS indices are dropped for simplicity

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 6

 Local Scattering Spatial Correlation Model (1)
Develop a model for R ∈ CM×M for a NLoS channel between a UE and a BS equipped with ULA
UE and BS indices are dropped for simplicity

Received signal at BS is sum of Npath multipath components, where Npath is a large number.

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 6

 Local Scattering Spatial Correlation Model (1)
Develop a model for R ∈ CM×M for a NLoS channel between a UE and a BS equipped with ULA
UE and BS indices are dropped for simplicity

Received signal at BS is sum of Npath multipath components, where Npath is a large number.
Suppose scattering is localized around UE. BS is elevated and has no scatterers in its near-field

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 6

 Local Scattering Spatial Correlation Model (1)
Develop a model for R ∈ CM×M for a NLoS channel between a UE and a BS equipped with ULA
UE and BS indices are dropped for simplicity

Received signal at BS is sum of Npath multipath components, where Npath is a large number.
Suppose scattering is localized around UE. BS is elevated and has no scatterers in its near-field
ϕ̄ = ϕ + δ, where ϕ is a deterministic nominal angle and

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 6

 Local Scattering Spatial Correlation Model (1)
Develop a model for R ∈ CM×M for a NLoS channel between a UE and a BS equipped with ULA
UE and BS indices are dropped for simplicity

Received signal at BS is sum of Npath multipath components, where Npath is a large number.
Suppose scattering is localized around UE. BS is elevated and has no scatterers in its near-field
ϕ̄ = ϕ + δ, where ϕ is a deterministic nominal angle and
δ is a random deviation from the nominal angle with standard deviation σϕ

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 6

 Local Scattering Spatial Correlation Model (1)
Develop a model for R ∈ CM×M for a NLoS channel between a UE and a BS equipped with ULA
UE and BS indices are dropped for simplicity

Received signal at BS is sum of Npath multipath components, where Npath is a large number.
Suppose scattering is localized around UE. BS is elevated and has no scatterers in its near-field
ϕ̄ = ϕ + δ, where ϕ is a deterministic nominal angle and
δ is a random deviation from the nominal angle with standard deviation σϕ
2
Gaussian distributed δ ∼ N (0, σϕ ),

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 6

 Local Scattering Spatial Correlation Model (1)
Develop a model for R ∈ CM×M for a NLoS channel between a UE and a BS equipped with ULA
UE and BS indices are dropped for simplicity

Received signal at BS is sum of Npath multipath components, where Npath is a large number.
Suppose scattering is localized around UE. BS is elevated and has no scatterers in its near-field
ϕ̄ = ϕ + δ, where ϕ is a deterministic nominal angle and
δ is a random deviation from the nominal angle with standard deviation√σϕ
2
Gaussian distributed δ ∼ N (0, σϕ ),Laplacian distributed δ ∼ Lap(0, σϕ / 2),

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 6

 Local Scattering Spatial Correlation Model (1)
Develop a model for R ∈ CM×M for a NLoS channel between a UE and a BS equipped with ULA
UE and BS indices are dropped for simplicity

Received signal at BS is sum of Npath multipath components, where Npath is a large number.
Suppose scattering is localized around UE. BS is elevated and has no scatterers in its near-field
ϕ̄ = ϕ + δ, where ϕ is a deterministic nominal angle and
δ is a random deviation from the nominal angle with standard deviation√σϕ
2
Gaussian distributed δ ∼ N (0, σϕ ),Laplacian distributed δ ∼ Lap(0, σϕ / 2), Uniformily distributed
√ √
deviations δ ∼ U[− 3σϕ , 3σϕ ]

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 6

 Local Scattering Spatial Correlation Model (1)
Develop a model for R ∈ CM×M for a NLoS channel between a UE and a BS equipped with ULA
UE and BS indices are dropped for simplicity

Received signal at BS is sum of Npath multipath components, where Npath is a large number.
Suppose scattering is localized around UE. BS is elevated and has no scatterers in its near-field
ϕ̄ = ϕ + δ, where ϕ is a deterministic nominal angle and
δ is a random deviation from the nominal angle with standard deviation√σϕ
2
Gaussian distributed δ ∼ N (0, σϕ ),Laplacian distributed δ ∼ Lap(0, σϕ / 2), Uniformily distributed
√ √
deviations δ ∼ U[− 3σϕ , 3σϕ ]
Standard deviation σϕ is measured in radians and is called the angular standard deviation (ASD)

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 6

 Local Scattering Spatial Correlation Model (1)
Develop a model for R ∈ CM×M for a NLoS channel between a UE and a BS equipped with ULA
UE and BS indices are dropped for simplicity

Received signal at BS is sum of Npath multipath components, where Npath is a large number.
Suppose scattering is localized around UE. BS is elevated and has no scatterers in its near-field
ϕ̄ = ϕ + δ, where ϕ is a deterministic nominal angle and
δ is a random deviation from the nominal angle with standard deviation√σϕ
2
Gaussian distributed δ ∼ N (0, σϕ ),Laplacian distributed δ ∼ Lap(0, σϕ / 2), Uniformily distributed
√ √
deviations δ ∼ U[− 3σϕ , 3σϕ ]
Standard deviation σϕ is measured in radians and is called the angular standard deviation (ASD)
since it determines how large the deviations from the nominal angle are

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 6

 Local Scattering Spatial Correlation Model (1)
Develop a model for R ∈ CM×M for a NLoS channel between a UE and a BS equipped with ULA
UE and BS indices are dropped for simplicity

Received signal at BS is sum of Npath multipath components, where Npath is a large number.
Suppose scattering is localized around UE. BS is elevated and has no scatterers in its near-field
ϕ̄ = ϕ + δ, where ϕ is a deterministic nominal angle and
δ is a random deviation from the nominal angle with standard deviation√σϕ
2
Gaussian distributed δ ∼ N (0, σϕ ),Laplacian distributed δ ∼ Lap(0, σϕ / 2), Uniformily distributed
√ √
deviations δ ∼ U[− 3σϕ , 3σϕ ]
Standard deviation σϕ is measured in radians and is called the angular standard deviation (ASD)
since it determines how large the deviations from the nominal angle are
A reasonable value of σϕ in urban cellular networks is 10o
MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 6
 Local Scattering Spatial Correlation Model (2)

Each multipath components results in a plane wave that reaches array from a particular angle ϕ̄n

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 7

 Local Scattering Spatial Correlation Model (2)

Each multipath components results in a plane wave that reaches array from a particular angle ϕ̄n
h iT
Array response of each path an = gn 1 e 2πj dH sin(ϕ̄n ) · · · e 2πj dH (M−1) sin(ϕ̄n )

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 7

 Local Scattering Spatial Correlation Model (2)

Each multipath components results in a plane wave that reaches array from a particular angle ϕ̄n
h iT
Array response of each path an = gn 1 e 2πj dH sin(ϕ̄n ) · · · e 2πj dH (M−1) sin(ϕ̄n )
gn accounts for gain of nth path. dH is antenna spacing

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 7

 Local Scattering Spatial Correlation Model (2)

Each multipath components results in a plane wave that reaches array from a particular angle ϕ̄n
h iT
Array response of each path an = gn 1 e 2πj dH sin(ϕ̄n ) · · · e 2πj dH (M−1) sin(ϕ̄n )
gn accounts for gain of nth path. dH is antenna spacing
gn are i.i.d. random variables with zero-mean and variance E{|gn |2 }.

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 7

 Local Scattering Spatial Correlation Model (2)

Each multipath components results in a plane wave that reaches array from a particular angle ϕ̄n
h iT
Array response of each path an = gn 1 e 2πj dH sin(ϕ̄n ) · · · e 2πj dH (M−1) sin(ϕ̄n )
gn accounts for gain of nth path. dH is antenna spacing
gn are i.i.d. random variables with zero-mean and variance E{|gn |2 }.
variance represents the average gain of the nth path

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 7

 Local Scattering Spatial Correlation Model (2)

Each multipath components results in a plane wave that reaches array from a particular angle ϕ̄n
h iT
Array response of each path an = gn 1 e 2πj dH sin(ϕ̄n ) · · · e 2πj dH (M−1) sin(ϕ̄n )
gn accounts for gain of nth path. dH is antenna spacing
gn are i.i.d. random variables with zero-mean and variance E{|gn |2 }.
variance represents the average gain of the nth path
PNpath
total average gain of the multipath components is β = n=1
E{|gn |2 }

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 7

 Local Scattering Spatial Correlation Model (2)

Each multipath components results in a plane wave that reaches array from a particular angle ϕ̄n
h iT
Array response of each path an = gn 1 e 2πj dH sin(ϕ̄n ) · · · e 2πj dH (M−1) sin(ϕ̄n )
gn accounts for gain of nth path. dH is antenna spacing
gn are i.i.d. random variables with zero-mean and variance E{|gn |2 }.
variance represents the average gain of the nth path
PNpath
total average gain of the multipath components is β = n=1
E{|gn |2 }
assume angles ϕ̄n are i.i.d. random variables with angular pdf f (ϕ̄)

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 7

 Local Scattering Spatial Correlation Model (2)

Each multipath components results in a plane wave that reaches array from a particular angle ϕ̄n
h iT
Array response of each path an = gn 1 e 2πj dH sin(ϕ̄n ) · · · e 2πj dH (M−1) sin(ϕ̄n )
gn accounts for gain of nth path. dH is antenna spacing
gn are i.i.d. random variables with zero-mean and variance E{|gn |2 }.
variance represents the average gain of the nth path
PNpath
total average gain of the multipath components is β = n=1
E{|gn |2 }
assume angles ϕ̄n are i.i.d. random variables with angular pdf f (ϕ̄)
Channel response Npath
X
h= an with
n=1

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 7

 Local Scattering Spatial Correlation Model (2)

Each multipath components results in a plane wave that reaches array from a particular angle ϕ̄n
h iT
Array response of each path an = gn 1 e 2πj dH sin(ϕ̄n ) · · · e 2πj dH (M−1) sin(ϕ̄n )
gn accounts for gain of nth path. dH is antenna spacing
gn are i.i.d. random variables with zero-mean and variance E{|gn |2 }.
variance represents the average gain of the nth path
PNpath
total average gain of the multipath components is β = n=1
E{|gn |2 }
assume angles ϕ̄n are i.i.d. random variables with angular pdf f (ϕ̄)
Channel response Npath
X
an with R = E hhH

h=
n=1

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 7

 Local Scattering Spatial Correlation Model (2)

Each multipath components results in a plane wave that reaches array from a particular angle ϕ̄n
h iT
Array response of each path an = gn 1 e 2πj dH sin(ϕ̄n ) · · · e 2πj dH (M−1) sin(ϕ̄n )
gn accounts for gain of nth path. dH is antenna spacing
gn are i.i.d. random variables with zero-mean and variance E{|gn |2 }.
variance represents the average gain of the nth path
PNpath
total average gain of the multipath components is β = n=1
E{|gn |2 }
assume angles ϕ̄n are i.i.d. random variables with angular pdf f (ϕ̄)
Channel response
 
Npath NX
path 
X
H
an aH

h= an with R = E hh =E n
 
n=1 n=1

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 7

 Local Scattering Spatial Correlation Model (2)

Each multipath components results in a plane wave that reaches array from a particular angle ϕ̄n
h iT
Array response of each path an = gn 1 e 2πj dH sin(ϕ̄n ) · · · e 2πj dH (M−1) sin(ϕ̄n )
gn accounts for gain of nth path. dH is antenna spacing
gn are i.i.d. random variables with zero-mean and variance E{|gn |2 }.
variance represents the average gain of the nth path
PNpath
total average gain of the multipath components is β = n=1
E{|gn |2 }
assume angles ϕ̄n are i.i.d. random variables with angular pdf f (ϕ̄)
Channel response
 
Npath NX
path 
X
H
an aH

h= an with R = E hh =E n
 
n=1 n=1

Multidimensional central limit theorem – as Npath → ∞, h → NC (0M , R)

MIMO Wireless Communications (Rohit Budhiraja, IITK) Correlation 7