Fundamentals of MIMO Wireless Communication

Prof. Suvra Sekhar Das

Department of Electronics and Communication Engineering
Indian Institute of Technology, Kharagpur

Lecture - 13
Coherence Time

Welcome to the course on fundamentals of MIMO wireless communications. We are

currently studying correlation properties of the received signal, when continuous wave is
transmitted and what is received what we have seen randomly fluctuating amplitude we have
seen is distribution, and we have also derived the expression for the correlation of I
component.

(Refer Slide Time: 00:48)

So, what we have last derived is the autocorrelation of hI hI with a lag Δt is:

So, this is where we had stopped our last discussion.

(Refer Slide Time: 01:18)

And to complete the description we also need to write the expression for now,

and hence the

or in other words the correlation of the received signal is completely characterized by the
correlation of the I components of the signal.
(Refer Slide Time: 02:38)

Now, this Bessel function as we have derived in this would like the one that is present in this
particular picture. So, this is the picture and we have also briefly shown you this particular
picture, and it is clear now that what we have this picture I was telling you in last time should
be little bit modified, because the picture is taken from book Stuber directly. So, this axis this

x axis is going to be f m Δt and this axis is as you are seeing is because is the
power that is normalizing this particular term. So, this correlation function as we have seen is
decreasing from a peak value 1 of towards 0 and because of the assumptions that we have
made we are getting a damp oscillation about 0. So, there might be negative values, but what
is of interest I mean of course, it is a model. So, there are problems which come with a
particular model. So, what would be interested fundamentally is in this range where the
correlation the coefficient drops to very small values, and since this axis is plotted with f m Δt
we can use this particular figure for any values for instance if you change f m , because

So, as we keep changing f c or as we keep changing v we will get different values of f m . So,
what we see over is as f m This is becoming smaller and smaller. So, a particular value of
coherence, this value of coherence appears at much-much smaller duration of f m Δt or in
other words if f m is large; that means, if you look at this product f m Δt . So, let us read 1
point if you take this line straight and we reach this point, which is let us say 70% . So, what
we can read this as at this value of f m Δt which let us say the correlation is 0.7 or

we could also state that . So, since , suppose we increase f m where

Δt must decrease in order to maintain the same value of x, as to get the same value of
correlation and vice-versa.

So; that means, as the Doppler frequency increases; that means, as the velocity increases, the
time at which there is 70% correlation would be smaller and smaller. So, what does it mean
this effectively means that as we increase our velocity, the time over which the channel is
correlated to itself is becoming less. So, if we read this curve in another way, any point over
here let us say this is the 0.9, let us say this point is 0.9 and we start going down this line
here. So, this much this duration this much duration of f m Δt corresponds to that argument
which would lead to 0.9 correlations. Now if I divide this by f m what I am left with is Δt .
So, Δt at which correlation coefficient is 0.9 is known as Δt0.9 = 90% coherence time.

Similarly, if I take this point and go down let us say point here, in this point as a value over
here. So, we would have f m Δt′ which is another value of Δt here because the same f m
divided by f m would be Δt′ and for this case we would call it Δt0.5 ; that means, it is the
separation in time which would lead to 50% coherence between single components. What it
means effectively is that, if we take the time axis this t axis and if we measure it in a two
instants t1 and t2 and in this gap between in this time is t2 − t1 = Δt right.

So, if this Δt = Δt0.5 ; that means, we are having we are at this point the signal would be 50%
correlated and what we have made assumption over here it is wide sense stationary; that
means, it is not dependent on t1 and t2 , but it is simply dependent upon the lag time between
t1 and t2 . So, this is one of the very, very important things in the propagation characteristics
which one should understand very nicely. So, I would ask you to spend some more time on
this, maybe spend time with the references that we have already mentioned relook at it and
try to understand it yourself again because, based on this the design of violence of
communication system is done significantly.
(Refer Slide Time: 09:33)

For instance it is, we would say that coherence time is the time for which the . So,
basically we are talking about this Δt , is equal to a certain value let us say 0.9 or 0.7 or 0.5
and so on and so forth. So, these are very, very important and remember the result we have

got is because we have made one fundamental assumption . If you

check these assumptions the results which we have obtained would be different. So, this

particular derivation that we made there was G(θ) p(θ) In this where over here
and we made G(θ) = 1 for which we have arrived at this expression and then we said if we
would change those assumptions this results would be different. So, whatever we are
discussing is with respect to these particular assumptions any other assumptions would lead
to different results.

So, what we are essentially trying to say is coherence time is the time over which signal is
correlated with itself if this is the time axis this is the h(t) signal what received would be
fluctuating right like this. So, that is what we have seen and we have been trying to get
and this is what we have been trying to capture. So; that means, we are taking
reading at this point and we are taking reading at some Δt some increase Δt and so on. And
we have arrived at this particular expression. So, when we say Δt0.9 so; that means we are
talking about separation in time over which the signals are 90% correlated.
So, in this figure we would probably have this separation which is probably as small as this
where the signals are correlated with itself, right. If I have a point here and a point here is
definitely the signals are not correlated because we do not know at this gap there is
uncorrelation. So, this duration of gap at which signals become uncorrelated would be a loose
definition of coherence time or the time separation at which signal becomes uncorrelated. So,

there is a rule of thumb for coherence time T c which is mentioned as . So, if you
use this particular formula you would get a rough calculation of the coherence time for a

particular signal. Little bit stricter definition of coherence time would be This is
again a rule of thumb.

But what is ideally to be done to calculate the coherence time is we should use this particular
picture that is the coherence time relationship and as we over here f m Δt and we should say
that suppose I want 90% coherence time. So, I would go to 90% I would read this particular
point and I would try to come to the x axis, once I have read this off I know this is x let us

say. So, I would take for any other distribution; that means, instead of taking the

pi as we have this particular expression, as we have in this expression in this

particular expression over here instead of if we do something else in that case the
curve would be different in that case we would not get the Bessel function in that case we
would not get the Bessel function it would be something else and if you are looking at 0.9
coherence we have to read up this 0.9 read this x axis from this x axis we have to find what is
the coherence time corresponding to the amount of coherence that we require.

This amount of coherence that we require could be 90% could be 70% could be 50%
whatever is the requirement. Effectively what do you mean by coherence time? Is the
separation in time for which the signals are correlated to itself. So, beyond this coherence
time the signal would not be correlated to itself. So, if we take two samples for instance I take
a sample here and I take a sample here they are not correlated. If I take a sample here they are
not correlated, if I take a sample at this point and I take a sample at this point they are not
correlated. Whereas when I keep the time interval small they are correlated now on the same
figure, if I would draw the situation where f m is very small; that means, slowly fluctuating.
So, in this slowly fluctuating case probably the channel would vary in this fashion. In that
case the coherence time would be large; that means 90% coherence time would be achieved
in this separation. So, here it could be 95% coherence so; that means, if you look at a very
small separation distance or it could be 95% correlated to itself. Whereas if you look at an
even larger distance we may find 90% coherence we may find 90% coherence in a larger
distance. So, this again an average value this is again an average value. Because this axis that
we are looking at . So, what we should remember is, at every instance
of separation of this Δt we may not get 90% correlation, but on an average if the separation
is this T c then you are going to get similarity between the values.

(Refer Slide Time: 16:15)

So, with this we go into a very important definition of our study. If suppose we have T s as
the symbol duration. Symbol duration means we talk about digital communication systems.
So, in a digital communication system we are sending symbols which are like this and so on.
So, let this be the symbol duration. So, if the coherence time as we measured in multiple
ways is larger than the symbol duration. So, what will happen?

The channel at this point and this point would be correlated. Because coherence time is larger
than symbol duration; that means, across this time this time is less than coherence time; that
means, the channel is nearly flat nearly equal to this. So, the basically could be fluctuating in
this fashion or in this picture, if you look corresponding to the red line the symbol duration
could be as small as this symbol duration could be as small as this. So, if we have symbols
going like this then that is the situation that we are talking about, if we have this situation
then what we have is known as slow fading. So, we had identified the channel to be time
selected in time selective if this is the condition then we say it is the slow fading condition.

Where as if coherence time is smaller than symbol duration; that means, the signal fluctuates
even before the symbol is over; that means, the channel is fluctuating like this, in that case it
is known as a fast fading situation all right. So; that means, once we calculate our coherence
time, once again let me remind you we could calculate coherence time using this formula
which is a little bit relaxed formula or this particular formula which is a little bit stricter. I
would recommend you to always use this formula if nothing is mentioned this is very, very
important if nothing is mentioned. If we use this you are having better margin and your
system design would be better. if with this basically the distance in time approach with signal
is uncorrelated.

So, that is also, but it is not that tight design, where is an exact design would be using the
derivation of correlation function for the corresponding p(θ) and from that you plot the curve
and you read off the 90% coherence point going to the x axis from which you calculate in the
time difference. So, from these three different ways you could calculate the coherence time
then you compare with the symbol duration once you find coherence time is larger than
symbol duration, you know it is a slow fading condition. if it is smaller than symbol duration
it is fast fading.

Now, why is it important? Because clearly if the coherence time is larger than symbol
duration; that means, we have this particular scenario, where the channel is fluctuating
slowly. So, the channel is fluctuating slowly could be because of two possibilities; one the
symbol duration is really, really small in the other is the mobility is very, very slow in either
of these two cases we would get a slow fading condition. Because we are doing relative
measurement, we are doing it in comparison to comparing the symbol duration with respect
to coherence time.

So, suppose symbol duration is 1 millisecond let us say symbol duration is 1 millisecond and
coherence time is 2 milliseconds. So, if 2 milliseconds are greater than 1 millisecond it is
slow fading right now you consider that this one millisecond is remaining the same. So,
suppose this is one millisecond and this is 2 milliseconds. Now using the same 1 millisecond,
if there is an increase in mobility; that means things start to move faster if things move to
faster the f m goes up. So, if f m goes up; that means, you are shrinking the x axis, if that
happens then the coherence time reduces. If coherence time reduces and as we increase the
mobility, if coherence time becomes comparable or less than the symbol duration you end up
in a fast fading situation.

Now, when you reach a fast fading situation; that means, let us say this has become 0.5
millisecond. Now when you are in the fast fading condition; that means, signal amplitude
fluctuates faster than the symbol duration what is going to happen. Suppose I have sent x
there is of course, noise at the receiver and what we said is fast fading condition h gets
multiplied. So, in a slow fading condition h would remain the same throughout the symbol
duration. So, I would receive the symbol y if I could estimate h then I could equalize the

effect of h for instance and I have able to recover x; that means, I would divide y
by h and I would getting back an estimate of x where as if the channel is fluctuating like this;

that means, if I have . so; that means, I am assuming two

different kinds of two different kinds of notable value of h; one over the period of 0 to t1 the
other of the period of t1 to T that is the symbol duration in that case we have an advance
equalizer, sometimes known as fraction space equalizer which would sample at a higher rate
and would try to do fractional equalization or even more complicated procedures.

So, typically we would like to design systems which would experience a slow fading
condition rather than a fast fading condition. If you cannot help creating a situation of slow
fading you have to design advanced receivers which would be able to handle the fast
fluctuation of signal strength.

So, what we saw in the example is that even though the symbol duration remains the same as
mobility, increasing coherence time has become smaller and then the system from slow
fading as now is experiencing fast fading. On the contrary if we say the earlier situation
where coherence time is 2 millisecond, I have a system which as band which as a symbol
duration of 1 millisecond, now suppose I use a different communication system I use a
communication system where symbol duration is much-much larger; that means, the symbol
duration is instead now let us say the 5 millisecond. So, when symbol duration is 5
millisecond all though mobility is giving 2 millisecond of coherence time it is experiencing
fast fading.

So, what we see is that the definition of slow fading or fast fading is a comparison of
coherence time with respect to symbol bandwidth that means, if I am transmitting from some
information let us take the example GSM and LTE, these two systems have different symbol
durations. So, these two systems will experience a different channel condition; one could
experience slow fading other could experience fast fading for the same mobility condition.
The coherence time as you have seen here in this particular expression is affected by f m
which is the Doppler frequency this f m is a function of velocity as well as it is a function of
carrier frequency f c . So, basically ; that means, as I increase v , f m increases as f m

increase, T c decreases right because clearly . As I increase velocity coherence

time decreases as I increase carrier frequency coherence time decreases right. So, as we
increase mobility coherence time would decrease.

On the other hand if I change by symbol duration again although the coherence time could
remain the same we would experience different kinds of fading condition.

(Refer Slide Time: 25:25)

So, with this we move on to see what happens to other properties of the signal. So, what we
would lastly like to see is the envelope correlation. Envelope correlation is basically the

correlation of where, .

(Refer Slide Time: 25:46)

So, the result that we have is what we are going to share with you over here, again the
derivation is cumbersome. So, we would not go through that, if you want you can go through
it. So, this particular expression, now this results are taken from Stuber I have used the

description from Stover. So, basically we have is . So, what you

see the difference between the previous one and the current one is that it is J 0 2 instead of
J 0 ; that means, it is decreasing faster, typical example of this would be the figure as shown
one over here. So, if you see this particular figure this is the envelope correlation.
(Refer Slide Time: 27:11)

So, this is decreasing faster than the other one.

(Refer Slide Time: 27:23)

And if you look at this square envelope correlation the square envelope correlation also gives

us the same thing that this is .

Now, the importance of you can clearly guess because . Now h(t)

for , . So, basically this gives us signals strength right. So, when we study
the correlation of We are basically studying the correlation of the signal strength. So,
what we have over here is that correlation of the signal is proportional to whereas
the correlation of the envelope alpha or the correlation of the alpha squared are both

proportional to .

So, this is under line thing should remember in this particular discussion and just to
summarize what we have also talk about in today’s discussion is the calculation of coherence
time we, have explained what is the meaning of a correlation function and how this
correlation is effecting and we have also explained rule of thumb for calculating the
coherence time for the special case that we are taking the special case is defined by

and finally, we have also described the fast and slow fading conditions,
where we have said that the fast fading condition is when the channel fluctuates faster than
the symbol duration and slow fading condition is when the channel fluctuates slower than the
symbol duration.

So, it is quite easy to remember when the signal fades at a rate which is slower than the
symbol rate it is a slow fading condition. Whereas when the signal fades at a rate which faster
than the symbol rate it is a fast fading condition, and with most of a analysis in this particular
course we will be mostly assuming a slow fading condition; that means, within the symbol
duration the channel is nearly constant and it is not fluctuating. However, from symbol to
symbol there is fluctuation of signal strength with time.

Thank you.