You are on page 1of 10

MIMO- ZF and ZF-IC

Equalization
REF: J. R. Hampton, “Introduction to MIMO Communications”, Cambridge University Press
MIMO System equation normalization
Different authors express the MIMO system equation in different form.

Normalization
𝑟ҧ = 𝛼𝐻 𝑠ҧ + 𝑧 𝛼 = constant 𝛼𝐻 = physical channel matrix

𝑅𝑟𝑟 = 𝐄 𝑟ҧ 𝑟ҧ H 𝐻 = normalized channel matrix

= 𝛼 2 𝜎𝑠2 𝐄 𝐻𝐻H + 𝜎𝑧2 I𝑁𝑟


Assumption: 𝐻 and 𝑠ҧ are independent. Transmitted signals have equal power from each transmit antenna and are
uncorrelated. 𝜎𝑠2 is the variance of each transmitted signal. 𝜎𝑧2 variance of the noise at the receive antenna.

𝑃𝑟𝑡𝑜𝑡 = Tr 𝑅𝑟𝑟 = 𝛼 2 𝜎𝑠2 𝐄 | 𝐻 |2𝐹 + 𝑁𝑟 𝜎𝑧2

𝑃𝑟𝑡𝑜𝑡 𝛼 2 𝜎𝑠2 𝐄 | 𝐻 |2𝐹


𝑃𝑟 = = + 𝜎𝑧2
𝑁𝑟 𝑁𝑟

𝛼 2 𝜎𝑠2 𝐄 | 𝐻 |2𝐹
SNR = =𝜌
𝑁𝑟 𝜎𝑧2
Let 𝛼 = 𝜌

𝜎𝑠2 𝐄 | 𝐻 |2𝐹
1=
𝑁𝑟 𝜎𝑧2

This equation can be satisfied in a number of ways.


Our normalization assumes that
2
𝐄 ℎ𝑖𝑗 =1 𝐻 is random

𝐄 | 𝐻 |2𝐹 = 𝑁𝑟 𝑁𝑡
1
Then if 𝜎𝑧2 = 1 and 𝜎𝑠2 = 𝑁 , then
𝑡

1 𝑁𝑟 𝑁𝑡
𝜌= 𝛼2 = 𝛼2
𝑁𝑡 𝑁𝑟 ∙ 1
𝛼= 𝜌

𝑟ҧ = 𝜌𝐻𝑠ҧ + 𝑧ҧ
ZF Equalization
−1
𝐻+ = 𝐻H 𝐻 𝐻H
Moore-Penrose pseudo inverse

We have a 𝑁𝑟 × 𝑁𝑡 MIMO system. During symbol period 𝑘


𝑁𝑡

𝑟𝑖 𝑘 = 𝜌 ෍ ℎ𝑖𝑗 𝑠𝑗 (𝑘) + 𝑧𝑖 (𝑘) 𝑖 = 1,2, … , 𝑁𝑟


𝑗=1

𝜌 is the SNR at the receiver. Let 𝑘 = 1,2, … , 𝑝, there will be 𝑝𝑁𝑟 equations.

𝑅 = 𝜌𝐻𝑆 + 𝑍 𝐻 = 𝑁𝑟 × 𝑁𝑡 channel matrix

𝑟1 (1) 𝑟1 (2) . . 𝑟1 (𝑝) 𝑠1 (1) 𝑠1 (2) . . 𝑠1 (𝑝) 𝑧1 (1) 𝑧1 (2) . . 𝑧1 (𝑝)


𝑟2 (1) 𝑟2 (2) . . 𝑟2 (𝑝) 𝑠2 (1) 𝑠2 (2) . . 𝑠2 (𝑝) 𝑧2 (1) 𝑧2 (2) . . 𝑧2 (𝑝)
𝑅= . . . . . 𝑆= . . . . . 𝑍= . . . . .
. . . . . . . . . . . . . . .
𝑟𝑁𝑟 (1) 𝑟𝑁𝑟 (2) . . 𝑟𝑁𝑟 (𝑝) 𝑠𝑁𝑡 (1) 𝑠𝑁𝑡 (2) . . 𝑠𝑁𝑡 (𝑝) 𝑧𝑁𝑟 (1) 𝑧𝑁𝑟 (2) . . 𝑧𝑁𝑟 (𝑝)
While doing spatial multiplexing, if the receiver 𝑖 attempts to detect the signal from a specific transmitter say 𝑙
then signals other than 𝑙 received at 𝑖 constitutes the interference.
𝑁𝑡

I𝑙 = 𝜌 ෍ ℎ𝑖𝑗 𝑠𝑗 (𝑘) 𝑘 = 1,2, … , 𝑝


𝑗≠𝑙
The signal-to-interference-plus-noise ratio at each receiver, SINR, is equal to
𝜌
SINR =
𝜌 𝑁𝑡 − 1 + 𝑁𝑡

−1
𝑅෨ = 𝐻 + 𝑅 = 𝐻H 𝐻 𝐻H 𝑅 In the case of zero-forcing, the effect
−1 of interference is reduced by pre-
= 𝐻H 𝐻 𝐻H 𝜌𝐻𝑆 + 𝑍 multiplying the received signal
−1 −1 matrix by the Moore–Penrose
= 𝜌 𝐻H 𝐻 𝐻H 𝐻 𝑆 + 𝐻H 𝐻 𝐻H 𝑍 pseudo inverse
= 𝜌𝑆 + 𝑍ҧ

If we denote 𝑗 th row of 𝑅෨ by 𝑟෥𝑗


𝑟෥𝑗 = 𝜌𝑠ഥ𝑗 + 𝑧෥𝑗
The last equation shows that 𝑗 th row of 𝑅෨ consists of a vector of transmitted symbols from the 𝑗 th transmitter and
has a noise vector but with no interference.

If we consider the 𝑘 th element of 𝑟෥𝑗 as 𝑟෥𝑗 (𝑘) We have shown that pre-
𝑟෥𝑗 (𝑘) = 𝜌𝑠𝑗 (𝑘) + 𝑧෥𝑗 (𝑘) multiplying R by 𝐻 + eliminates
interference, or, equivalently,
𝑟𝑗 𝑘 − 𝜌𝑠𝑗 (𝑘)|2
𝑠ෝ𝑗 𝑘 = arg min |෥ that it forces the interference to
𝑠𝑗 (𝑘) 𝑘 = 1,2, … , 𝑝
be zero; hence, the terminology
zero-forcing. In this process,
however, the noise term has also
ZF with SIC been pre-multiplied by 𝐻+ ,
resulting in a modified
QR factorization of 𝐻, noise matrix 𝑍.ҧ Unfortunately, the
𝑚 × 𝑛 matrix 𝑚 ≥ 𝑛 can be decomposed into elimination of interference in
𝑄𝑉 𝑄 = unitary matrix of 𝑚 × 𝑚
zero-forcing detection
𝑉 = upper triangular 𝑚 × 𝑛 comes at the expense of a growth
in noise power called noise
𝐻 = 𝑄𝑉 amplification.
𝑅෨ = 𝑄H 𝑅 = 𝜌𝑄H 𝐻𝑆 + 𝑄H 𝑍
= 𝜌𝑄H 𝑄𝑉𝑆 + 𝑄H 𝑍 = 𝜌𝑉𝑆 + 𝑄H 𝑍

𝑁𝑟 × 𝑁𝑡 matrix 𝑁𝑟 ≥ 𝑁𝑡
𝑣11 𝑣12 . 𝑣1𝑁𝑡
0 𝑣21 . 𝑣2𝑁𝑡 𝑠1 (1) . . 𝑠1 (𝑝) 𝑧෥1 (1) . . 𝑧෥1 (𝑝)
. .
𝑅෨ = 𝜌 . .
. 𝑣𝑁𝑡𝑁𝑡
. . . . + . . . .
0. 0. 𝑠𝑁𝑡 (1) . . 𝑠𝑁𝑡 (𝑝) 𝑧෦ . .
. . 𝑁𝑟 (1) 𝑧෦𝑁𝑟 (𝑝)
0 0 . 0

𝑖 th row at 𝑅෨
𝑁𝑡

𝑟𝑖ǁ 𝑘 = 𝜌 ෍ 𝑣𝑖𝑗 𝑠𝑗 (𝑘) + 𝑧ǁ𝑖 (𝑘)


𝑗=1

𝑁𝑡

= 𝜌 𝑣𝑖𝑖 𝑠𝑖 𝑘 + ෍ 𝑣𝑖𝑗 𝑠𝑗 (𝑘) + 𝑧ǁ𝑖 (𝑘)


𝑗=𝑖+1
Let 𝑖 = 𝑁𝑡
𝑟𝑁
ǁ 𝑡 𝑘 = 𝜌𝑣𝑁𝑡𝑁𝑡 𝑠𝑁𝑡 (𝑘) + 𝑧ǁ𝑁𝑡 (𝑘)

𝑧ǁ𝑁𝑡 (𝑘)
𝑟𝑁
Ƹ 𝑡 𝑘 = 𝑠𝑁𝑡 (𝑘) + 𝑘 = 1,2, … , 𝑝
𝜌𝑣𝑁𝑡 𝑁𝑡

Ƹ 𝑡 𝑘 − 𝑠𝑁𝑡 (𝑘)|2
𝑠𝑁𝑡 (𝑘) = arg min |𝑟𝑁
𝑠𝑁𝑡 (𝑘)

Estimating the symbols transmitted by antenna 𝑁𝑡 during a frame of length 𝑝 symbols.


Next, 𝑖 = 𝑁𝑡 − 1

𝑟𝑁
ǁ 𝑡−1 𝑘 = 𝜌 𝑣(𝑁𝑡−1)(𝑁𝑡−1) 𝑠𝑁𝑡 −1 𝑘 + 𝑣(𝑁𝑡−1)𝑁𝑡 𝑠𝑁𝑡 (𝑘) + 𝑧ǁ𝑁𝑡−1 (𝑘)

where 𝑣(𝑁𝑡−1)𝑁𝑡 𝑠𝑁𝑡 (𝑘) is the interference from 𝑁𝑡 .

Since we have estimated signals from layer 𝑁𝑡 , these estimates are used to cancel interference from layer 𝑁𝑡 .
Substract 𝜌𝑣(𝑁𝑡 −1)𝑁𝑡 𝑠Ƹ𝑁𝑡 (𝑘) from 𝑟𝑁
ǁ 𝑡−1 𝑘

𝑟𝑁Ӗ 𝑡−1 𝑘 = 𝑟𝑁
ǁ 𝑡−1 𝑘 − 𝜌𝑣(𝑁𝑡−1)𝑁𝑡 𝑠Ƹ𝑁𝑡 (𝑘)

= 𝜌 𝑣(𝑁𝑡−1)(𝑁𝑡−1) 𝑠𝑁𝑡 −1 𝑘 + 𝑣(𝑁𝑡−1)𝑁𝑡 𝑠𝑁𝑡 𝑘 − 𝑠Ƹ𝑁𝑡 (𝑘) + 𝑧ǁ𝑁𝑡−1 (𝑘)

𝑟𝑁Ӗ 𝑡−1 𝑘
𝑟𝑁
Ƹ 𝑡−1 𝑘 =
𝜌𝑣(𝑁𝑡−1)(𝑁𝑡−1)

𝑣(𝑁𝑡−1)𝑁𝑡 𝑠𝑁𝑡 𝑘 − 𝑠Ƹ𝑁𝑡 (𝑘) 𝑧ǁ𝑁𝑡 −1 (𝑘)


= 𝑠𝑁𝑡 −1 𝑘 + +
𝑣(𝑁𝑡−1)(𝑁𝑡−1) 𝜌𝑣(𝑁𝑡 −1)(𝑁𝑡 −1)

𝑣(𝑁𝑡 −1)𝑁𝑡 𝑠𝑁𝑡 𝑘 −𝑠Ƹ 𝑁𝑡 (𝑘)


where is the residual interference from layer 𝑁𝑡 .
𝑣(𝑁𝑡 −1)(𝑁𝑡 −1)

Since the interference cancellation has been performed using the estimate of signal from layer 𝑁𝑡 , the impact of
interference from layer 𝑁𝑡 is reduced.
We can perform maximum likelihood detection on 𝑟𝑁
Ƹ 𝑡−1 𝑘 to obtain the estimate of symbols from layer 𝑁𝑡 − 1,
i.e. 𝑠Ƹ𝑁𝑡 −1 𝑘

Generalizing

σ𝑁
𝑗=𝑖+1 𝑣𝑖𝑗 𝑠𝑗 𝑘 − 𝑠𝑗Ƹ (𝑘)
𝑡
𝑧Ƹ𝑖 (𝑘)
𝑟𝑖Ƹ 𝑘 = 𝑠𝑖 𝑘 + +
𝑣𝑖𝑖 𝜌𝑣𝑖𝑖

You might also like