Lecture 45 : 5G sounding reference signal (SRS) design – part I

IIT Kanpur July 201
20 followers
3 Video Views·Sep 13, 2024
Okay, welcome to the class. We have been discussing this DMRS. We will finish discussing this
DMRS generation, okay and I promise that we will discuss it in, finish discussing in the
last class. We could not finish this. We will just finish it now and then we will start
discussing this sound and reference signal generation, how they are generated in the
standard, okay. So this is what we saw. I will quickly run through this.
So first this reference symbol QPSK modulated reference symbols are designed, okay and later
on they are multiplied with this orthogonal code cover sequences and this slide and the
next slide is telling us how these orthogonal code cover sequences are applied on to the,
on to these reference symbols, QPSK modulated reference symbols, okay. So I showed you this
formula, okay and basically this is the amplification factor. These are the orthogonal covers, okay,
which are used to finally generate the orthogonal cover which is used to cover, to create that
orthogonal covering, okay, right and it is a generic formula which is used for both type
1 and type 2, okay and so this is L dash is 0 for single symbol, 0, 1 for double symbol.
These sequences are given in table below, okay. We saw that if you look at directly
at this, it does not make sense, but if you substitute these values over here, it will
be clear what is happening, okay and that is what I am going to show you immediately,
okay. So, but one thing is clear that this table is meant for type 1, okay, wherein like
you see this maximum 8 ports can be defined, okay, right and these different groups, but
how many groups are there, okay, only 2 groups, CDM group 0 and CDM group 1, okay.
So if you have to go to type 2, there should be 3 different CDM groups, okay, correct and
so that is, this is, this is limited to the type 1, okay. This formula is applicable for
both type 1 and type 2, WF and WT which are coming over here. See F is frequency, other
one is time, okay. So, we will see that how this is used, okay.
I take up this example, okay, wherein this, what is this? This is double symbol, okay
and what is this? This is type 1 double symbol, okay. How many CDM groups are there? Group
0 and group 1, okay, 2 different groups. That is what is shown in the last table, okay,
right. So, you see that this is much better seen now, okay and what is happening over
here? So, let us look at this AKL, okay and this is the port, okay. So, what is coming
over here? Let us, if you substitute everything over here, okay, so it will be clear what
is happening, okay. So, if you look at this, look at this, let us try to basically see
what is happening.
L dash is 0 for single symbol, 0 and 1 for double symbol, okay. What is L bar? L bar
is basically the starting of the DMRS, okay. So, this one, what will this index tell,
okay, whether you are in the first symbol, this is the starting, okay, whether you are
in the first symbol or in the second symbol. L dash is 0, it means that you are looking
at this symbol. L dash is 1, what is that? You are looking at this symbol. What is this
K dash and 0 and 1? You are looking at these subcarriers, right, okay, correct. That is
what is happening, okay, and this is some running index, okay, correct.
So, if you have to, so basically what are we saying? Here we have to generate this orthogonal
code cover, okay, for the CDM group 0, okay, that is what it is, okay, when the port 0,
0, 0, okay. Now, if you look at this, you substitute K dash 0, okay, and L dash 0, what
are you going to get, okay, multiply. So, what is happening? You are multiplying WF
K dash 0 and WF L dash 0, okay, which position is going to generate? You are going to generate
this position, right, okay. So, what is happening? K dash 0 and L dash 0, product of them is
positive, right, that is what is coming over here, okay.
Then you increment, you can increment. So, if you want to go to the time domain, what
will you have to increment? You will have to increment L dash because L dash was, will
be 1 for the second symbol, that is what I said in the last slide, okay, correct. So,
if you make L dash as 1, what will happen over here? You are moving over here, okay.
So, what is this? L dash is 1, you multiply it with K dash is equal to 0 because this
is the first subcarrier, correct, okay. So, what am I saying here? To generate this, this
particular value which is positive, okay, how will you generate it? You have to look
at this, okay. What are you doing here? What is this formula telling you? This is the amplification
factor, okay. What is this? K dash is 0, N dash is, N is 0, that is the running index,
okay. What will happen? It will be R of 0. You take the first symbol, okay, correct and
what are you doing? Then you will multiply it with the product of these two, correct,
okay and the product of these two for this position, what is that coming? It is positive.
So, you take this symbol and multiply it with a positive value. That is the first part,
first bit of that 4-bit orthogonal code cover, okay, right. Now, you want to see this, what
is this value? What will you do? You will increment L dash. You will keep K dash as
same, okay, because you are still looking at the first subcarrier, right, okay.
So, what is happening? You see this K dash is 0, L dash is 1. You again get positive.
That is what is coming over here, right, okay. So, you take the, so you have incremented
K dash to 1. You get the first, N is still 0, right. So, what will happen? You will have
the first, this was 0, okay. So, you will get the first value of the DMRS. Multiply
it with plus 1. That is what you did for the first time. You did this, okay. The second
one also, the second symbol, what will you do? You will have to multiply it with the
plus 1 because that is the sequence, this multiplied with this, right, okay.
Look at this now, okay. What happens? K dash, you will have to move to the first one because
you are incrementing along the frequency, right. What will happen? K dash is 1. You
look at L dash is 0. It is again positive. Multiply this, okay, right. You take the third,
so what will happen over here? You will move on to 0, 1. You will move on to the second
one, okay. Second value, this 0, 1 and basically the third value, the index by second, it will
get multiplied by plus 1 again. That is what is happening. Fourth one will also, you will
get this plus 1 and plus 1. That is why this, that is how this 4 bit orthogonal code cover
is created, right. So, 0 gets multiplied, R of 0 gets multiplied by plus, R of 1 also
gets multiplied by plus, R of, again R of 2 gets multiplied by plus and R of 3 will
get multiplied by, okay. Now you move, if this is our code, what happens?
For this code, you see this, plus minus, okay. So, K dash 0 and L dash 0 will give you plus
1 and plus 1. That is what is coming over here, right, okay. And here K dash 0 and L
dash 1, again plus 1 and plus 1. So, that is what is coming over here, plus plus, right.
Is it clear what am I saying here, okay? If you move along time for this port, correct,
so this, how will you move along time in this table? Take K dash is equal to 0, move from
L dash. So, this is the first bit, which is this. What will be the second bit? You move
along time, okay, from L dash is equal to 0 to L dash is equal to 1. What is happening?
Plus 1 and again gets multiplied by plus 1, right. So, this is the second, fine, okay.
What happens to the third one, which is coming over the second subcarrier, okay? What happens?
K dash becomes minus 1. Multiply minus 1 with plus 1. This is minus 1. Multiply and then
you have to move along time, okay. So, you will move from here to here. You will, earlier
you are multiplying K dash is equal to 1 with L dash is equal to 0. You got minus 1 and
then you move from L dash is equal to 0 to L dash is equal to 1 to check for this. Again
minus, right, okay. So, that is how it is created, okay. That is how it is defined in
the standard. This is the formula, okay, right. So, you look at this. So, however, this will
actually tell you, okay, that if K dash for this type, for type 1 configuration, okay,
this delta is basically come over here, okay. Delta is 0, okay. You will have to move on
by K dash is equal to 0. So, you will hop from one to other. That is what it is telling
you, okay. You are looking at this for configuration type 1. You are hopping from, you are hopping
every even times, okay, right. That is the expression which is coming over here, okay.
So, that is how you generate. I have given you only broadly how it is generated, okay,
right. So, this is how the orthogonal code cover is generated, okay. So, this is, I took
an example for type 1 double symbol, okay. Similar table should be there for type 2,
for type 2 also, okay, right. So, this brings us to the end of basically this DMRS generation.
Similar, you can take an example for type 2 also and can do that as a homework exercise,
okay. So, it is, the logic is same, okay, fine. So, this is how DMRS are defined in
this SRS generation, okay. So, similarly, see the SRS also needs to be generated for
different number of ports, okay, right and you see that they need to be generated for
different users also, okay. So, we will see that how that is done, okay.
So, what are we doing here, okay? Basically, before we get into the details of understanding
SRS, it is good to recapture why SRS is even required, okay. Forget MIMO, it is like even
if basically user has a single antenna, okay, right, then and the base station has to pre-code
the PDSCH, okay. It should know this channel, okay. It should know the downlink channel,
okay, correct and how does it estimate this downlink channel? The use case which we have
been considering is basically the TDD based beamforming, correct, okay. So, what happens?
So, these are the two problems which GNV faces before it can pre-code PDSCH, okay. GNV needs
to estimate the channel and know the MCS, okay. We have seen this, okay. No point in
spending too much of time over here. So, we, what we are saying here is GNV operates in TDD
mode, okay, where it can easily infer downlink channel from uplink channel. This is an assumption,
okay, right. We have discussed this in detail, okay. So, what are this concept of SRS?
Before estimating the uplink channel, you will transmit the pilot X to the G node B,
okay. Pilot is called the sounding reference signal, okay and how does that happen? This
is, we are calling this as a pilot, okay. So, you see this, the single antenna, right. How
many layers can be transmitted? What is the rank of the channel? Rank of the channel will be 1,
okay, correct, okay and now how many ports you need for this user? You need only a single SRS
port, right, okay and so that is what is being transmitted. This is a simple example wherein
we are doing it over a single subcarrier, correct and this subcarrier, this pilot is
transmitted and it is received here, okay, over two antennas and I am not even talking
on multiple SRS ports. This is only single SRS port, okay, correct and so this SRS is transmitted.
Then, you get for our first antenna, you get y1 is equal to h1x plus n1. These are two system
of the equations, okay. So, y1 is h1 plus n1, y2 is h2 plus n2, correct, okay. It gets multiplied
by x and what have we done here when from here to here. So, this is the actual received signal
over two antennas. When writing from here to here, I have set x as 1, okay, just for the sake
of simplicity, okay. So, these are the channel estimates and this is an SRS, okay, right.
So, in this case, we are only talking about a single port SRS, okay and I am going to discuss
in this class how this different SRS are defined. That is all, okay and this is one of the examples
which where we have used the SRS, okay. So, there are other examples wherein like SRS are used,
okay. Now, this is basically the example which we have been discussing in this case,
in our class. So, downlink channel acquisition assuming TDD reciprocity.
There are other uses of SRS which we have not discussed, initial scheduling, okay, right. So,
the user enters the network, right, okay and so base station does not know its channel,
okay. It is a 100 megahertz wide channel and each user will have a different channel. Now,
base station has to estimate, has to find out where, how is this particular user channel who
has just entered the network. How does it come to know? How does it estimate? So, basically assigns
this SRS to the user. User sends this SRS to the base station and base station is able to find out
how this like 100 megahertz channel or a part of 100 megahertz channel looks like for this user,
okay, correct and that is a very important concept, okay.
That see there is a 100 megahertz channel, okay and there are different kind of users in the
system, okay. Let us say there is a IoT user, okay, a massive machine type communication user.
This is operating in a very narrow bandwidth path, right, okay. As we discussed earlier,
like 5G is designed for different kind of users and then some of these users,
they have very tight energy requirements like a water meter, okay. It will not be
operating in 100 megahertz. It will be operating in very small bandwidth path.
So, for this user, the base station will like to find out a channel in very narrow bandwidth path
when compared with a like high bandwidth user, okay. So, the idea is that base station has to
find out the channel of each and every user before it can schedule this user, right, okay.
That is what I am saying here. This SRS is used for initial scheduling also,
okay. Forget this downlink channel acquisition. It comes later, okay. Even before I have to do
simple scheduling, I have to have this initial estimate of a channel, okay, right and then the
final use is basically the way we use, you remember CSIRS for the downlink to calculate
the MCS information, right, okay. So, if this TDD reciprocity is applicable in the downlink,
it is applicable in the uplink as well, okay, right. So, what happens here? So,
UE estimates the downlink channel using CSIRS, okay, right and since the channel is reciprocal,
okay, it knows the uplink channel under reciprocity assumption.
Now, UE designs the precoders for the uplink, not the base station, okay, and it precodes
this SRS and then base station using this precoded SRS, okay, calculates the uplink MCS
and then reports it back to the user so that it can use that particular MCS for the uplink.
You get that, okay. What are we saying here? This is exactly opposite of what we did for the
downlink, okay. In the downlink, what happened, okay? Base station used the precoded CSIRS. Here,
it is using the, UE is using the precoded SRS, okay, for the base station to find out
the equivalent MCS for the uplink, right, okay. So, that is another use case which we never
touched upon. There is not a need, but this is also done, okay. So, these are three broad use
cases of SRS, okay, right and now I will show you how it is defined in the standard, okay, fine.
Look at this, okay. So, SRS, first thing is, okay, unlike DMRS, okay, it is limited to a
maximum of 4 antenna ports, okay, right and there is a reason, okay, why this is limited to 4
antenna ports, why it is in the downlink, there are maximum of 12 DMRS ports, okay.
UE is basically limited by the, uplink is limited by the UE power,
okay and so typically UE power is on a lower side than the base station power,
okay, right. So, you cannot, it is difficult to support more than 4 layers in the uplink.
That is what people have realized, okay, with all the uplink constraints, okay. Let me put it this
way, okay. With all the downlink constraint, people have done their simulations and research
to figure out that maximum 12 layers can be supported in the downlink, but in the uplink,
only 4 layers can be supported. So, you see this, there is a difference in the way SRS is designed,
okay, only 4 antenna ports, okay, right and so what is this? This is the time symbol duration,
okay. This is n SRS and the symbol is the number of symbols over which the SRS is defined,
okay, right. So, SRS time duration can be 1 symbol, you look at this, correct or 2 or 4
consecutive symbols within a slot. So, here it is 4 consecutive symbols.
They are always consecutive symbols, okay and you will see here, so the starting symbol location is
denoted as L0, something similar to what we saw for DMRS also, okay and starting symbol,
you see this here, where it is coming? Towards the end, okay. So, what is this telling? What is
this bullet telling us? That the starting symbol can be located anywhere within the last 6 OFDM
symbol, okay, of a slot provided the resource, resource does not cross the end of the slot
boundary. What is it saying? And this SRS has to lie within this slot, okay. That is how typically
we always transmit, okay, but what is this? This is to find, so these are some aspects,
okay. What is, what are we saying here? That SRS is defined only for a maximum of 4
antenna ports, okay and they typically come at the end of the slot, okay. Why? First, if you see that
what is, what happens in the first part of the slot is typically allocated for the downlink,
right, okay. This is the first 2 symbols is PDCCS transmission, next few symbols will be
PDSCS transmission, okay. If it is a basically the TDD system, then you switch
only towards the end of a slot, correct, okay. So, that and that part is only the uplink
transmission. That is why you see this SRS coming only at the end, okay. Here, it is coming in the
last 4 symbols. Here, it is coming in the second last symbol, but you look at this bullet. What
is it saying? It can come anywhere within the last 6 OFDM symbols. You cannot say that, okay,
it will come in this last 6 symbols because it will cross the slot boundary, okay. So,
you have to follow that basic rule, okay, then SRS should not cross the slot boundary, okay.
Now, so what is this? You should get used to this terminology. SRS is actually a QAM structure.
What is this QAM? You look at this, okay. Here, it is easier to see, okay. These are the brushes
of the QAM, okay, right and that is why it is called a QAM structure, okay, typically coming
at the end, okay. This end, if you see, it looks like a QAM, okay. It is not coming somewhere in
the middle, okay and so QAM spacing is either 2 symbols, okay. They are separated by basically
this. So, KTC is, so QAM spacing, okay, is either 2 or 4, not symbols, is either 2 or 4,
okay and what are we saying? It is either separated by 2 resource elements or is either
or separated by 4 resource elements. That is what this bullet is telling, okay,
fine and then this is a symbol which is used in a standard to decide whether this QAM spacing
is 2 or 4, KTC, okay. Then, what does this technically mean, okay? What does this QAM
spacing mean? It means that SRS is transmitted on every nth subcarrier where n can be 2 or 4.
If QAM spacing is 2, the SRS will be transmitted on this every second subcarrier, okay and if the
QAM spacing is 4, it will be transmitted on every fourth subcarrier, fine, okay.
So, what is this QAM spacing allowing us to do? SRS structure allows frequency domain
multiplexing of different SRS of different UEs, okay, right. So, what is happening over here?
So, how, why it is designed like this, okay? As I said, for each UE, base station has to
find out the channel. Let us say it is initial scheduling, okay. A user is entering the network,
okay and the base station has to sound out. That is the term which is technically used.
Base station has to sound out the channel of each user, okay. Basically, figure out
how good or bad this particular bandwidth part of each user is, okay. So, we will assign,
let us say, this QAM, okay, for this particular user, for a particular user, okay, right and
in the standard, if it has to like find out the channel of second user,
we will assign the second QAM, okay, right. So, I have shown it for one PRB, okay. This is,
we will extend it for, let us say, 275 PRBs, okay. This QAM can be defined over 275 PRBs,
okay. So, by using this kind of a QAM, base station will be able to find out the channel of,
the entire channel of UE1 and then what it will do? Know the channel over these subcarriers,
over this it will interpolate, correct, okay and similarly for the second user,
it will assign the other QAM, which will be offset by this one symbol, correct, okay, right
and so it will be able to find out basically the complete, let us say, 100 megahertz channel,
just for an example, okay, right and for different users, okay.
Now, this is, how many users can be multiplexed with this kind of QAM structure? Only two,
right. How many can be, how many UEs can be simultaneously sounded out with this
QAM2 over one symbol? Only two, right. One will be assigned for this, for UE1 and then offset
and it will be assigned to the UE2, correct, okay. Is this clear what I am saying here that
QAM2 structure in this case can allow you to multiplex only two users, okay. That is why
different QAM structure is defined, okay. So, different QAMs are offset by each other by
integer number of Re's, okay. So, here they are offset by 1, here they are offset by 4,
right, okay. So, what is happening here? If this is, now look at this, look at this QAM structure,
okay. What happens here? This QAM structure is basically, let us say it is for one user,
okay, correct. Then for the second user, I can offset it by 1, okay, right and I can assign
it to the second user. Similarly, I can do for another two users, right. So, this kind of QAM
structure is allowing me to multiplex 4 different users and base station can find out the entire
frequency domain channel of all these 4 users, right, simultaneously. Assign, I have to like
sound out 4 users and do this allocation, okay. Correct, if I have to sound out 2 users, I will
do this allocation, okay. And this, if you look at this, why this is done? Basically, this also
we are going to discuss in detail, okay. You see this, there are like 4 adjacent symbols,
okay. What, why is this being done? Let us say UV is in a poor SNR condition,
correct, okay and then the channel does not change much, okay. So, this because the channel
remains constant, then the channel can be basically averaged to improve the SNR, right,
okay. So, that is why one reason wherein like you see this kind of QAM structure.
So, all 4 contiguous, it can be 2 also, okay and it can be 4 depending on how bad or good
the UV channel condition or SNR conditions are, okay. But it allows to basically sound out 4
different users and so this is something called the QAM offset, kappa bar TEC, which is defined
with respect to a PRB boundary, okay. So, this, I will cover these aspects when we talk about this,
this frequency hopping, okay. So, there is this QAM offset which is defined,
okay, with respect to a PRB boundary, okay and where is this, like where is this QAM coming,
okay, that is what it typically tells that, okay and there is a repetition factor also associated.
Once we start talking about the frequency hopping, that is how this is done. These 2 aspects I
mentioned, okay, just for the record, their use will become clear when we start talking about this
frequency hopping structure. I will take an example, okay, to basically tell you that how
this QAM structures are used for different users, okay, right. So, look at this. So, this is
physically, so QAM spacing and offset example. So, this, that was the example I was giving you,
okay, right. So, this is the QAM offset. This is one QAM for you being used for user blue.
This is another QAM being used for user green, right, okay, correct. That is what we said,
okay, that for this kind of QAM, you can multiplex 2 different users, right, okay.
So, that is what is happening, okay. Now, so and here this QAM spacing is how much? QAM spacing
is 2, right, okay and here you look at it. How much is the QAM spacing? QAM spacing is 4,
right, okay and what are you saying here? SRS with a QAM spacing of 4 has a lesser density
than QAM spacing of 2, okay. So, this, what kind of, how many like SRS you can design,
how many users you can multiplex with this kind of QAM spacing? Look at this, okay. Here,
every alternate RE is being used as SRS. Here, every fourth RE is being used as SRS, okay,
but there is no repetition along time, okay, right. So, we have taken this QAM structure,
okay, further reduced it, okay, to allow standard allows that, okay.
So, what is happening here? This, these 2 users, okay, so this QAM is allowed for user 1. Here,
this QAM is allowed for user 2, right. This is another user, user 3. A separate QAM is defined,
okay. Here, so every fourth RE is used as a SRS, okay. This is user yellow. For that,
again a separate QAM is defined, okay. So, in for this kind of QAM, how many users I can
multiplex? I can multiplex 4 different users. I have shown only 2 over here, okay, but I can
multiplex 2 more users and sound out their channel also as well, okay.
How have I been able to achieve this? Basically, reducing the number of REs,
which I have used for defining the SRS, okay, SRS QAM. Here, every alternate is being used. So,
RE is being used. Here, every fourth RE is being used for one QAM, okay. So, this is QAM 1. This
is second QAM. QAM third and QAM fourth are not being used, okay. I can use that also, okay,
right. So, standard gives this kind of flexibility in designing the QAM, which will help in
basically sounding out the channel of different users simultaneously, okay, right, because what
is happening here? By allocating one QAM to user 1 and another QAM to user 2, all these, I am able,
as a base station, I am able to figure out or estimate channel of these 2 users in the same
symbol itself, right, an entire, let us say, 275 PRB channel, which is 100 megahertz channel,
okay. So, this is one. Here also, same thing is possible, but with the lower density, right. So,
here you will get better interpolation. Here you will get the poorer interpolation. It depends on
what kind of user this is. If it is a, let us say, IoT user, which does not require this kind of
good channel estimates, where the data rate is low, this kind of pilot assignment could,
SRS assignment could be done. If it is basically high data rate user, where the channel estimates
have to be extremely good, right, you will have to do this kind of assignment or application based.
If it is a voice application, right, it is okay, if like if the estimates are poor,
then receiver quality, receive signal quality can be slightly poorer than
wherein it is a URLC user, okay, wherein you cannot afford to have a poor quality channel
estimate because the reliability is important, latency is important, okay. So, it is a function
of all that, okay. So, standard gives us this kind of flexibility. We will continue discussing
this in the next class also. Thank you.
