The capacity can be composed by decomposing the vector channel into a set of parallel independent scalar Gaussian sub-channel.
Using SVD ::
H = U Λ V*
U and V are uniary matrix
Λ is a rectangular matrix whose diagonal elements are non-negative and the others are zero.
|λ1 |
| λ2 |
| λ3 |
| λmin |
λi*λi is the eigen values of H.H*
u is eigen vector of H.H*
v is the eigen vector of H*.H
y = ((U Λ V*). x)+n
Let y’ = U* . y
y’ = U*. ((U Λ V*). x)+n
= Λ V* . x + U*.n (lets put V*. x = x’ and U*. n = n’)
= y’ = Λ . x’ + n’ (where Λ => (λ1 >= λ2 >= λ3 >= … >= λmin)
We can represent this operation using the parallel sub-channels ::
(yi)’ = λi . (Xi)’ + (ni)’ , (i = 1,2,3…. nmin)
X’—>(multiplier: λ1)—>(adder: n1′)—->y1′
. . . .
. . . .
. . . .
X<nmin>’—>(multiplier: λ<nmin>)—>(adder: n<nmin>’)—-> y<nmin>’
If full CSI ,
Introduce waterfilling scheme to MIMO ::
Pi* = (μ – [N0/(λi*λi)] ) ^ t with μ chosen to satisfy the total power constraint
Summation(i=0 to <nmin>) Pi* = P
C = Summation(i=1 to <nmin>) log ( 1+ ( [Pi*.(λi*λi)]/N0) //like , seriously improved channel.
—————————————————————————
So , after doing all of this – improving the capcity and all that – we also need to consider –
The parameters that determine the performance.
Ofcourse – the best place to start is with SNR (high or low)
1.High SNR – equal power allocation has almost the same performance as waterfilling. Therefore in each subchannel , the
transmit capacity (considering k = nmin) :: P / k
total C = summation(i=1:k) log(1+ P/k * (λi*λi)/N0)
This k is basically minimum of (nt,nr)<— number of spatial degrees of freedom.
in case of high SNR , we ignore the 1 in the C equation , so we get ::
C = summation(i=1:k) log(P/k * (λi*λi)/N0) further simplifying…
C= summation(i=1:k) log(P/N0) + summation(i=1:k) log(λi*λi)/k) …
but we know .. log(P/N0) = SNR
SO… C= summation(i=1:k) log SNR(Cawgn for high SNR) + summation(i=1:k) log(λi*λi)/k) :: HIGH SNR
So.. A MIMO channel provides Nmin spatial degrees of freedom.
If we go ahead and divide C by k , we get this whole thing where the summation pretty much turns into an expectation E of log(1 + [P.(λi*λi) / k.N0])
Then if we go ahead and use the Jensen’s Inequality (cuz log is a Concave function)
E(f(x)) < f(E{x})
C/k = log {1 + P/k.No * (1/k * summation(i=1 to k) (λi*λi) ) }
where … [summation(i=1 to k) (λi*λi)] = Trace [ H.H*] = summation(i,j) { |h<i,j>|*|h<i,j>| }
So , the total power gain of the matrix channel among the channels with the same total power gain , the one which has the higher capacity is the one with all the singular values equal.
The condition number of H is the ratio of => max λi : min λi => 1 cuz when all the singular values are equal , min λi = max λi = 1 . Sooo… —> H is said to be well-conditioned and also in this case we have the highest capacity.
Well-Conditioned channel matrix facilitate communication in the high SNR regime.
Phew… so much for high SNR. What about LOW SNR ?
Using the Waterfilling scheme to allocate power to the highest (λi * λi) (that is the best channel)
Best Channel = max of (λi*λi).
Best Channel Capacity = C = log ( 1 + [ P . max<i>(λi*λi) ] / N0)
C ~ P/N0 . max<i>(λi*λi) . log(base 2) e ….. cuz [ log (1+X) ~ X log(base 2) e at small X )
C~ { max<i>(λi*λi) . log(base 2) e }. SNR
Fast Fading : L. Tc (time diversity)
MIMO : Spatial Diversity
C > Cawgn
Want to read more? : Read here
More insights to come….
![](https://i0.wp.com/cdn.memegenerator.net/instances/500x/47664431.jpg)
This Channel