Fundamentals of MIMO Wireless Communication

Prof. Suvra Sekhar Das

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

Lecture – 17
FSF -Coherence Bandwidth, Delay Doppler Characteristics

Welcome to the lecture on fundamentals of MIMO Wireless Communications, we have

currently discussing frequency selective fading we have already seen how frequency
selective fading Fourier transform or the transfer function has to be found out. We will
now look at the frequency characteristics of the signal and that will describe or give us a
full description of frequency selectivity, and we will also try to define coherence
bandwidth which is again another important parameter called such a Channel Propagation
Model. With that we will come to a close of this small scale propagation models for a
single input, single output system and a after that we need to proceed towards the MIMO
channels.

(Refer Slide Time: 01:01)

So, this is the model what we have started looking at, we have described this particular
model with the previous lecture.
(Refer Slide Time: 01:10)

We have also seen how the channel impulse response looks like this is our standard
classical description of channel impulse response, we have seen the phase what we have
added to the description is now that description is a little more details, where we have said
that well the first delay; first delay as given by this is also a some all the delays of all the
paths that come at the first delay plus all the paths coming at the second delay and these
two are getting separated because of delta function.

Since τ1 ≠ τ2 , we cannot add them up. So, here also that delta function separation was
available. So, this is what we have described in the previous lecture.
(Refer Slide Time: 01:53)

We want to see the frequency domain representations of we want to take the Fourier
transform in a delay domain. And what we have is the discrete Fourier transform is what
we are taking in this case I have also describe we could take the continue Fourier transform
also, and this is how it would look like and the expression of the Fourier transform is the
one that we have here. So, this is the expression that we have tine correction here would
be τ1 and τ2 , this is this is the right over here this is this is k over here.

So, what we have here is basically the Fourier transform of the echoes coming at the first
delay plus Fourier transform of the once coming at the second delay. So, what we last
discuss is if we write in the in the discrete domain let us look at in the discrete domain

itself this is 𝐻 (𝑡, 𝑘) = 𝐻(𝑡, 𝑘τ1 ) + 𝐻(𝑡, 𝑘τ2 ) + ⋯ + 𝐻(𝑡, 𝑘τ𝑛 ) + ⋯ + 𝐻 (𝑡, 𝑘τ𝑁𝑚𝑎𝑥 ); that

means, a maximum resolvable delays. So, we just remember there are two things one is 𝑁,
which is ideally to be 𝑁τ𝑛 . So, basically the 𝑁 we have used in this is ideally to be 𝑁𝜏𝑛 ;
that means, the number of multi paths at the 𝑛𝑡ℎ delay; that means, the first delay would
have 6 to 10 multi paths let say the second delay could have different number 1, 2, 3, 4, 5,
6, 7 this is second ellipse the third one could have 1, 2, 3, 4, 5, 6, 7, 8, 9, 10.

So, basically this is the third delay. So, what we can say is 𝑁τ1 = 10, 𝑁𝜏2 = 20, 𝑁𝜏3 = 15
and. So, this could be variation. So, we should keep in mind and this is the number of
resolvable delay. So, we could go up to 𝑁𝜏6 we could. So, this is basically indicated by
𝑁𝜏𝑚𝑎𝑥 .

So, if we proceed what we just said is if there is a single multi path we have already seen
it what would happen that would result in flat fading, but now we have addition of several
such once. So, what I will show you now is the frequency response picture of a sample
realization. So, remember this particular one that we have over here, yeah, this particular
one is a sample realization mind it at a function of time. So, if I say 𝐻 (𝑡 ′ , 𝑘) = ℎ(𝑡 ′ , τ1 )
where 𝑡 ′ ≠ 𝑡 for the first path, because there are different time instance. So, that we
understand. So, same would have to happen with the frequency selective fading. So, that
is why we say we are looking at a single snapshot proceed with this of course, we are given
this description.

(Refer Slide Time: 05:54)

(Refer Slide Time: 05:58)

So, now suppose I have taken one particular situation and this situation is described by
certain parameters which is not very much necessary at this point of time we can look at
them later through an example. So, when there is only one path or one resolvable path we
have seen it is flat across frequency. So, what you are going to see is this particular one
that is in red color across the frequency index. So, this is the frequency index clearly the x
axis with k axis frequency say and is the gain; that means, you are basically talking of
10 log10 |𝐻 (𝑡, 𝑘)|2 in dB. Whereas, if you describe it shortly we have taken bandwidth of
20MHz, this 50 nanoseconds as the spacing and RMS delay speed of 10 nanoseconds we
will describe more of this. So, this is what we get if we move further what we get is the
second situation where we say the delay spread has increased if we look at this the delay
spread has increased in this particular case.

So, I go back a little bit allow me. So, in the in the second case what we have is delay
spread from 10 nanoseconds has increased to 50 nanoseconds, what does that mean?
Previously suppose this is an echo if this is impulse that I have sent the delay spread is
very, very small delay and this width is 10 nanoseconds. Where is in the second case where
it is 50 this number is 50. So, if I clear it up if you look at this number here I take a different
color if you can if this is readable 10 and 50 the second one is 50 in that case let the first
echo come over here and then there are echoes.
So, this width is now 50 nanoseconds. So, how can this be 50 nanoseconds this can be 50
nanoseconds, if the reflectors or scatterers are spread in such a way that if this is the
transmitted this is the receiver, if this is an ellipse in the first case everything is almost-
almost here in the first case. In the second case reflector are slightly oriented in way there
are multi paths which are having longer delays. So, this is 50 nanoseconds right so; that
means, the spread is more, if the spread is more we are going to add up more resolvable
paths; that means, if we look at the previous picture; that means, we are getting more
resolvable paths, we are getting more resolvable paths as we are seeing this plus this, so
more, more and more.

So, this gets added if delay is less if delay is less you have only one you have flat if there
are a greater number of paths; that means, delays are more. More number of resolvable
delays if τ1 , 𝜏2 and so on. If this increases this spread increase. So, as this spread increases
what we get is the situation represented by this particular figure, by this plot that is the
frequency response that is happening. If we move on further and we say that we have a
possibility where I am erasing this for visibility and yes now we have seen – for, 100
nanoseconds which I will try to explain with different color; if it is 100 nanoseconds it
means that in the response is due to an impulse, impulse is launched at here is echoes are
coming in 100 nanoseconds; that means, echoes are coming for 100 nanoseconds.

So, I would say echo impulse launched the echoes are here. So, this gap is 10 nanosecond
the second case I would choose a different color and say that echoes are now spread to 50
nanoseconds, right in the third case which is indicated by the black one I would have
echoes which are coming from faraway places; that means, there are reflectors or scatterers
which are located far away. So, this distance is 100 nanosecond this means that, we have

more things to add up; that means, we will have what we have is ℱ (ℎ(𝑡, 𝑘τ1 )) +

ℱ (ℎ(𝑡, 𝑘τ2 )) and this for the red case is only limited to one for the blue case it is possibly

limited to 2 for the black case it is possibly limited to 3 or 4; that means, ℱ (ℎ(𝑡, 𝑘𝜏4 )), let

say and so on. And as we increase this length this line is now having tau the delay spread
as 1000 so; that means, the impulse if we could draw with let say yellow color.

Now, yellow is probably not nicely visible with this violet color let say. Echo starts here
echo skip coming for and it goes on for a duration when, this distance goes beyond this
particular screen is 1000 nanosecond so; that means, you will have several such things

getting added plus dot, dot, dot Fourier transform of ℱ (ℎ(𝑡, 𝑘𝜏20 )); that means, there are

20 resolvable multi path for example. So, what we observe is as we increase this delay,
what we are getting is a for the first case when there was a single resolvable delay this is
a flat fading the second case when there are let say 2 or little bit close 2 we have somewhat
variability in the frequency domain; that means, some white portion of the frequency is
getting selectively pass through while the others are not in the next case what we have is
this particular contour indicating that there is some portion of the frequency which is
getting signals not attenuated much where as there is another portion where the signal is
attenuated heavily.

The third case that we are going to see is or the forth case is when the signal is fluctuating
according to the line that I am facing screen has become very, very dirty. So, there is much
more fluctuation in the frequency domain, there is selectivity these shaded bands are the
ones which are having less attenuation, they are passing the signal properly where as this
region is having attenuation let say relative attenuation. So, what we see as we increase
the path. So, as we increase the delay the selectivity in the frequency range in the frequency
increases; that means, there is more crossing, there is a greater number of crossing if we
have a thresh holding the frequency axis there is a greater number of crossing and some
portions of the frequency axis this is the frequency axis are selectively allowing the signals
to pass through and hence this is known as frequency selective fading.

In this same tone, I would like to measure I would like to mention that the time selective
fading there it was instead of p instead of k that was the time axis and things looked very
similar and we had a time selective fading, in time selective fading what we remember we
had discussed certain portion of the time signals are passing through less attenuation
certain portion there was more attenuation and we describe level crossing rate and average
duration of fade. So, here also similar thing applies in the frequency domain and we move
quickly to the description.
(Refer Slide Time: 14:50)

So, along with this multi path power delay profile we would describe with something
known as the Coherence Bandwidth this is very, very important. So, coherence
bandwidth is described as the portion of bandwidth where the signal remains coherence
with itself just like we have defined coherence time now we are defining it in the frequency
domain; that means, single is coherent in the frequency domain, if I take to frequencies 𝑓1
and 𝑓2 over these two frequencies if the if the signal is quite same; that means, if
𝐻 (𝑡1 , 𝑘1 ) ≈ 𝐻 (𝑡1 , 𝑘2 ) that is what we have used k; that means, this 𝑘1 and 𝑘2 are within
the coherence bandwidth. We have a way of defining it as of now we will describe it very
briefly there are many detail descriptions to this also. So, what we clearly see in the
previous slide in the previous picture, what we have seen is as your number of paths
increase or your delay increases your frequency selectivity increases. So, as your delay
increases naturally the delay spread is going to increase.

So, as you delay spread is increasing the coherence of the signal across the bandwidth
decreases. So, if you take the first case coherence across this entire bandwidth is there. So,
coherence bandwidth is very large and delay spread is very, very small because there is
hardly spread in the delay. In the second case what we have is in the frequency domain
there is some more fluctuation delay is little bit more in the frequency domain the
coherence has reduced, if we take the black case then there is more delay; that means, it is
become 100 from previous case of 50 and 10.

The coherence in the frequency axis is less now that can be visibly seen over here and as
we move to the case of 1000, what we clearly see is that coherence bandwidth visibly it
has become less. So, one way of quantifying this is another way of describing the
frequency selectivity of the channel one is RMS delay spread the other is this particular
definition of coherence bandwidth, where coherence bandwidth is defined as inversely
1
propositional to RMS delays spread: 𝐵𝑐 ∝ τ , and there are two coherence bandwidth
𝑟𝑚𝑠

descriptions the 50% coherence bandwidth description and the 90% coherence bandwidth
is descriptions, these are exactly similar to the coherence time description the
philosophically they are the same.

1
So, the 50% coherence bandwidth is given by this expression one 𝐵𝑐 (0.5) = 5⋅τ and
𝑟𝑚𝑠

1
90% coherence bandwidth is 𝐵𝑐 (0.9) = 50⋅𝜏 ; that means, once we calculate the delay
𝑟𝑚𝑠

spread one divided by 50 times RMS delay spread is basically the portion of bandwidth
1
where signal is 90% correlated with each other. Whereas is the portion of the
5⋅𝜏𝑟𝑚𝑠

bandwidth where signal is 50% correlated with each other. So, two important definitions,
if we want the signal to be nearly constant across frequency we would find the bandwidth
given by 90% coherence bandwidth this 90% coherence bandwidth is across which the
signal is nearly flat. So, over this bandwidth you have flat fading you can say that is a good
approximation over the 50% coherence bandwidth there is some similarity, but there is
some variation also in the signal.

So, with RMS delay spread and with coherence bandwidth, we can describe the channel
characteristics of a frequency selective fading.
(Refer Slide Time: 18:47)

Ah just to support whatever we have discussed this particular slide contains some results
taken from papers which are from Donald Cocks and others. This particular picture
represents this particular picture shows us the x axis as delay spread so; that means, τ𝑟𝑚𝑠
and this axis is the coherence bandwidth that 50% coherence bandwidth see 0.5 correlation
is 50% coherence bandwidth. So, one is the measured versus some theoretical description
of it, what we clearly see is as the delay spread increases the coherence bandwidth
decreases.

So, what you see over is as see if you look at this portion delay spread is large coherence
bandwidth is small we large delay spread small coherence bandwidth. If we look at this
portion of the curve, delay spread is small coherence bandwidth is large this is matching
measurements with theoretical values. This is the 90% coherence bandwidth description
again it is quite similar what you say for large delay spread smaller coherence bandwidth
smaller delay spread a larger coherence bandwidth. So, that is as per description and
matching, matching with results this is for this 910MHz urban mobile radio channel; that
means, in the city area in 910MHz of center frequency.

So, what we see is that these descriptions are supported by measurements also. So, we are
actually following proper model, we are not without model finally, towards the end
because there are many things to discuss.

(Refer Slide Time: 20:27)

But we will directly go into something known as this scatter plot what we talk about in
this particular slide is the total description of the channel in the delay Doppler domain. So;
that means, we have seen the delay domain it is the τ axis see these are all τ axis 𝜏 is
mentioned over here this we have seen, the tau axis. What I had drawn previously was the
time axis; that means, I have drawn the time axis and I had said that signal would fluctuate
in time this is the 𝜏 axis, at another delay signal would fluctuate in time another delay
signal would fluctuate in time.

Now, we said that this delay could have 10 multi paths, this delay could have 20 this could
1
have 15 and so on and so forth, and also this delay could have a 𝑝(θ) = 2π. Where as this

one could be the one which has a specular component this one could have a different
1
𝑝( 𝜃 ) ≠ they could be different. Now 𝑝(𝜃) affects the Doppler that is what we have
2𝜋

seen because well calculating the Doppler we were calculating phi of h h of delta t. While
we calculated ϕℎℎ (Δ𝑡) we had 𝐸θ , now while having 𝐸𝜃 we made the assumption that let
1
𝐺 (θ); that means, antenna gain equals to one and 𝑝(𝜃) = 2𝜋 that is 2D isotropic scattering

model and there we said this gives raise Jakes spectrum that is what we have seen before.
So, here instead of this time variation which is fluctuating in time, if we look at the
Doppler’s spectrum which we expect to be the remaining static over a region and over
certain duration of time, because it is the second order statistics. So, it is not the
instantaneous coefficients.

So, this is one way of looking at the channel. So, in the 𝜏 axis we will take a look at the
average power in the in the frequency axis we will take look at this spectrum. So, what we
see in this particular picture which is the rural area diagram where it says at the first echo
the Doppler spectrum is a Jakes, along with it there is a specular component; that means,
it is rise it is the vision. At the second delay which is at 0.2 micro second we have Jakes
spectrum and so on and so forth. Whereas if we take look at an opposite one I mean these
are also similar this and these are similar. So, we take look at another one there are lot of
echoes coming in very short duration and after a very large duration if you at this is around
15 to 20 microseconds where as this is around 0.6 microsecond this is because it is a hilly
region there is hill across far away there is another hill. So, it takes long time for the signal
to get reflected and come back to the receiver.

So, mobile this is the transmitter right. So, in the in the reflectors which are close by
because there are let say houses around the mobile for example, things get scattered almost
quite well from all the directions what you have is a Jakes spectrum because in these cases
1
𝑝(𝜃) = 2𝜋. Whereas here the one that comes reflected from a far away reflector which
1
could be a hill, it is caused indicating it is 𝑝(𝜃) ≠ 2𝜋; that means, probably p of rays

coming from all direction is not equal the rays are coming from a specific direction and
the spectrum is given by the gauss spectrum which we had seen earlier. So, with this kind
of a diagram you can almost completely characterize the single input single output channel,
and look at how what would be the signal experiencing. There are many details to it, but
because of course, our interest we would limit ourselves to this without this description it
is difficult to proceed with the other things, but this is just the initial part of it there are
many more details which can be studied we would simply like to stop at this point with
the very short description that we have seen the small scale fading, where we have seen
time selective frequency selective in time.
(Refer Slide Time: 24:35)

We have seen slow fading and fast fading; slow fading is one where coherence time is
greater than symbol duration fast fading is one, where coherence time is small than smaller
than symbol duration. In frequency selective fading we have seen there are two cases one
is flat one is frequency selective, flat is the case which we could define as the coherence
bandwidth is larger than this system bandwidth and frequency selective is the case where
coherence bandwidth is smaller than the system bandwidth. Then is within the system
bandwidth if this is your system bandwidth, channel fluctuates in frequency and this is the
case within the system bandwidth the channel is nearly constant it is fluctuating later. Here
within the symbol duration channel is nearly constant here beyond the symbol duration or
within the symbol duration the signal is fluctuating.

With this we would bring our discussion on frequency selective and frequency flat fading
channels towards an end. There would be a certain numericals which we will try to provide
as additional material which would be useful for this particular part.

Thank you.